Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
METHODS FOR CLASSIFYING POPULATIONS INCLUDING ALZHEIMER'S DISEASE POPULATIONS
Document Type and Number:
WIPO Patent Application WO/2016/144838
Kind Code:
A1
Abstract:
The present disclosure describes methods for classifying two or more different populations, such as Alzheimer's disease. The disclosed method(s) utilizes a novel two-stage analysis for enhancing separation, e.g., biomarkers associated with Alzheimer's disease patients from age-matched controls and non-Alzheimer's disease demented patients. The analysis accounts for the linear dependence of output variables on input variables that are altered by, e.g., variables such as fetal bovine serum, age of the patients, and/or protein concentration.

Inventors:
CHIRILA, Florin, V. (895 Vandalia Road, Morgantown, WV, 26505, US)
ALKON, Daniel, L. (4604 Dorset Avenue, Chevy Chase, MD, 20815, US)
Application Number:
US2016/021101
Publication Date:
September 15, 2016
Filing Date:
March 06, 2016
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
BLANCHETTE ROCKEFELLER NEUROSCIENCES INSTITUTE (8 Medical Center Drive, Morgantown, WV, 26404-3409, US)
International Classes:
G01N33/50; G01N33/68; G06F17/10
Foreign References:
US20140248648A12014-09-04
US20140329264A12014-11-06
US20140248648A12014-09-04
US20140038186A12014-02-06
US7595167B22009-09-29
US20140031245A12014-01-30
US20110212474A12011-09-01
Other References:
CHIRILA FLORIN V ET AL: "Spatiotemporal Complexity of Fibroblast Networks Screens for Alzheimer's Disease", JOURNAL OF ALZHEIMER'S DISEASE, IOS PRESS, NL, vol. 33, no. 1, 1 January 2013 (2013-01-01), pages 165 - 176, XP009182993, ISSN: 1387-2877
CHIRILA FLORIN V ET AL: "Fibroblast Aggregation Rate Converges with Validated Peripheral Biomarkers for Alzheimer's Disease", JOURNAL OF ALZHEIMER'S DISEASE, IOS PRESS, NL, vol. 42, no. 4, 1 January 2014 (2014-01-01), pages 1279 - 1294, XP009182980, ISSN: 1387-2877, DOI: 10.3233/JAD-140672
F.V. CHIRILA ET AL., J. ALZHEIMER'S DISEASE, vol. 33, 2013, pages 165 - 176
F.V. CHIRILA ET AL., J. ALZHEIMER'S DISEASE, vol. 42, 2014, pages 1279 - 94
W.Q. ZHAO ET AL., NEUROBIOL. DIS., vol. 11, 2002, pages 166 - 183
T.K. KHAN ET AL., PROC. NATL. ACAD. SCI. U.S.A., vol. 103, no. 35, 2006, pages 13203 - 13207
T.K. KHAN ET AL., NEUROBIOL AGING, vol. 31, no. 6, 2008, pages 889 - 900
T. K. KHAN ET AL., NEUROBIOL. DIS., vol. 34, no. 2, 2009, pages 332 - 329
F.V. CHIRILA ET AL., J ALZHEIMER'S DISEASE, vol. 33, 2013, pages 165 - 176
D.W. JAYME ET AL., NATURE, vol. 334, 1988, pages 547 - 548
H. HANIU ET AL., TOXICOL IN VITRO, vol. 27, no. 6, 2013, pages 1679 - 85
E. FALKNER ET AL., TOXICOL IN VITRO, vol. 20, 2006, pages 395 - 400
D. BRUNNER ET AL., ALTEX, vol. 27, no. 1, 2010, pages 53 - 62
J. VAN DER VALK ET AL., TOXICOL IN VITRO., vol. 24, no. 4, 2010, pages 1053 - 1063
D. W. JAYME ET AL., CYTOTECHNOLOGY, vol. 33, no. 1-3, pages 27 - 36
H. HANIU ET AL., TOXICOL IN VITRO, vol. 27, no. 6, pages 1679 - 85
F.V. CHIRILA ET AL., J ALZHEIMER'S DISEASE, vol. 42, 2014, pages 1279 - 94
P.T. SPELLMAN ET AL., MO/EC. BIOL. OF THE CELL, vol. 9, 1998, pages 3273 - 3297
K. DE NARAY ET AL., APJS, vol. 165, 2006, pages 461 - 479
K. DE NARAY ET AL., APJS, vol. 676, 2008, pages 920 - 943
H.S.M. COXETER: "Introduction to Geometry", 1961, JOHN WILEY & SONS. INC.
I.M. YAGLOM: "Geometric Transformations", MATHEMATICAL ASSOCIATION OF AMERICA
D. ELIZONDO, IEEE TRANSACTIONS ON NEURAL NETWORKS, vol. 17, no. 2, 2006, pages 330 - 344
F. LAVIGNE ET AL., FRONT PSYCHOL., vol. 5, no. 842, 2014, pages 1 - 24
D.L. ALKON ET AL., BIOLOGICAL PLAUSIBILITY OF SYNAPTIC ASSOCIATIVE MEMORY MODELS, 1994
LEARNING AS SELF-ORGANIZATION, 2013, pages 247 - 259
D.L. ALKON ET AL., BIOL. CYBERN., vol. 62, 1990, pages 363 - 376
M. BLAIR ET AL., MEMORY AND COGNITION, vol. 29, no. 8, 2001, pages 1153 - 1164
R. HARALICK ET AL., PATTERN RECOGNITION, vol. 3587, 2005, pages 132 - 14
S. TAVAZOIE ET AL., NAT. GENET, vol. 22, no. 3, 1999, pages 281 - 285
P.T. SPELLMAN ET AL., MOLEC. BIOL. OF THE CELL, vol. 9, 1998, pages 3273 - 3297
A.L. YARBUS ET AL., BIOFIZIKA, vol. 6, no. 2, 1961, pages 52 - 56
K.T. BLACKWELL ET AL., J. EXPERIMENTAL & THEORETICAL ARTIFICIAL INTELLIGENCE, vol. 9, no. 4, 1997, pages 491 - 508
D.L. ALKON ET AL.: "Biological Plausibility of Synaptic Associative Memory Models", LEARNING AS SELF-ORGANIZATION, 1994, pages 247 - 259
J. B. TALCOTT ET AL., NEUROPSYCHOLOGIA, vol. 51, no. 3, 2013, pages 472 - 481
I.M. YAGLOM, GEOMETRIC TRANSFORMATIONS , , , MATHEMATICAL ASSOCIATION OF AMERICA
LEARNING AS SELF-ORGANIZATION, 2013
WHALLEY K.: "Neurodegenerative disease: Dishing up Alzheimer's disease", NATURE REVIEWS NEUROSCIENCE, vol. 13, March 2012 (2012-03-01), pages 149
WENDER ET AL., ORGANIC LETTERS (UNITED STATES, vol. 7, no. 10, 12 May 2005 (2005-05-12), pages 1995 - 8
WENDER ET AL., ORGANIC LETTERS (UNITED STATES, vol. 7, no. 6, 17 March 2005 (2005-03-17), pages 1177 - 80
WENDER ET AL., JOURNAL OF MEDICINAL CHEMISTRY, vol. 47, no. 26, 16 December 2004 (2004-12-16), pages 6638 - 44
CHIRILA F.V. ET AL., J. ALZHEIMER'S DISEASE, vol. 33, 2013, pages 165 - 176
SPELLMAN P.T ET AL.: "Comprehensive Identification of Cell Cycle-Regulated Genes of the Yeast Saccharomyces cervisiae by Microarray Hybridization", MOLE. BIOL. OF THE CELL, vol. 9, 1998, pages 3273 - 3297
KUZIO DE NARAY, T.; MCGAUGH, S.S.; DEBLOK, W.J.G.; BOSMA, A.: "High Resolution Optical Velocity Fields of 11 Low Surface Brightness Galaxies", APJS, vol. 165, 2008, pages 461 - 479
KUZIO DE NARAY, T.; MCGAUGH, S.S; DEBLOK, W.J.G.; BOSMA, A.: "Mass Models of Low Surface Brightness Galaxies with High Resolution Optical Velocity Fields", APJ, vol. 676, 2008, pages 920 - 943
Attorney, Agent or Firm:
BURGY, Adriana, L. (Finnegan, Henderson Farabow, Garrett & Dunner,901 New York Avenue, N.W, Washington DC, 20001, US)
Download PDF:
Claims:
WHAT IS CLAIMED:

1 . A method for classifying an Alzheimer's disease (AD) population from an age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing the biomarker assay on the biological samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

2. The method according to claim 1 , wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

3. The method according to claim 1 , wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

4. The method according to claim 3, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

5. The method according to claim 1 , wherein the samples comprise human skin fibroblast cells.

6. The method of claim 5, wherein the human skin fibroblast cells are cultured in fetal bovine serum.

7. The method of claim 5, wherein the biomarker is chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative biomarker (ADNI).

8. The method according to claim 7, wherein when the biomarker is fibroblast aggregation, the output variable is the natural logarithm of (Area/Number) and the input variable is the natural logarithm of cell density and fetal bovine serum.

9. The method according to claim 7, wherein when the biomarker is fractal dimension, the input variable is the intercept for the fractal curves and the output variable is the inverse slope.

10. The method according to claim 1 , further comprising establishing a universal cut-off value.

11 . The method according to claim 10, wherein the performing step (c) comprises establishing a discrimination limit having a cut-off value based on the signal-to-noise ratio, and wherein the plurality of data points generated from the AD population, AC population and/or the non-ADD population do not cross over the cutoff value.

12. A method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) obtaining a biological sample from the subject;

(b) performing a biomarker assay on the biological sample;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

13. The method according to claim 12, wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

14. The method according to claim 12, wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

15. The method according to claim 14, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

16. The method according to claim 12, wherein the samples comprise human skin fibroblast cells.

17. The method of claim 16, wherein the human skin fibroblast cells are cultured in fetal bovine serum.

18. The method of claim 16, wherein the biomarker is chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative biomarker (ADNI).

19. The method according to claim 18, wherein when the biomarker is fibroblast aggregation, the input variable is the natural logarithm of (Area/Number) and the output variable is fetal bovine serum.

20. The method according to claim 18, wherein when the biomarker is fractal dimension, the input variable is the intercept for the fractal curves and the output variable is the inverse slope.

21 . The method according to claim 12, further comprising establishing a universal cut-off value.

22. The method according to claim 21 , wherein the performing step (c) comprises establishing a discrimination limit based on the signal-to-noise ratio, and wherein the plurality of data points generated from the two or more AD populations do not cross over the cut-off value.

23. A method for classifying an Alzheimer's disease (AD) population from an age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative biomarker (ADNI) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

24. The method according to claim 23, wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

25. The method according to claim 23, wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

26. The method according to claim 25, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

27. A method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) providing skin fibroblast cells from the subject;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative biomarker (ADNI) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept; (e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

28. The method according to claim 27, wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

29. The method according to claim 27, wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

30. The method according to claim 29, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

31 . A method of establishing a ranked order for two or more Fetal Bovine Serum (FBS) lots using an assay with FBS comprising:

(a) obtaining biological samples from two or more different populations;

(b) performing the assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the two or more different populations are separated; and

(g) plotting a dynamic range of the two or more different populations, wherein the plot results in a ranked order of the two or more FBS lots.

32. The method according to claim 31 , wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

33. The method according to claim 31 , wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

34. The method according to claim 33, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

35. The method according to claim 31 , wherein the samples comprise human skin fibroblast cells.

36. The method of claim 31 , wherein the dynamic range is based on the distance between the two or more different populations.

37. The method of claim 31 , wherein the dynamic range is based on the average coefficient of variation normalized by the distance.

38. A method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot;

(b) performing the assay with the untested FBS lot;

(c) providing or generating a plurality of data points from two or more populations with the biological assay using two or more additional FBS lots, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the subjects of the two or more populations are separated;

(h) normalizing the biomarker result from the untested FBS lot;

(i) plotting a dynamic range of the two or more additional FBS lots and the untested FBS lot, wherein the plot results in a ranked order of the total FBS lots.

39. The method according to claim 38, wherein a first plurality of data points comprises controls and a second plurality of data points comprises Alzheimer's disease samples.

40. The method according to claim 38, wherein a first plurality of data points comprises controls, a second plurality of data points comprises Alzheimer's disease samples, and a third plurality of data points comprises non-Alzheimer's disease demented samples.

41 . The method according to claim 40, wherein the non-Alzheimer's disease demented samples are chosen from Huntington disease samples and Parkinson's disease samples.

42. The method according to claim 38, wherein the samples comprise human skin fibroblast cells.

43. The method of claim 38, wherein the dynamic range is based on the distance between the two or more different populations.

44. The method of claim 38, wherein the dynamic range is based on the average coefficient of variation normalized by the distance.

45. A method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot;

(b) performing the biological assay with the untested FBS lot;

(c) normalizing the biomarker result from the untested FBS lot;

(d) plotting the untested FBS lot normalized result on a ranked order of tested FBS lots.

46. The method according to claim 45, wherein the assay is a diagnostic assay.

47. The method according to claim 46, where the diagnostic assay is for Alzheimer's disease.

48. A method for classifying two or more different populations based on a diagnostic system comprising:

(a) obtaining samples for classification from the two or more different populations; (b) performing an assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input variable, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step (d) by translating each plurality of data points upward by the intercept b for each respective linear function;

wherein the method results in the separation of the two or more different populations.

49. The method according to claim 48, wherein the diagnostic system is chosen from Alzheimer's diagnostic assays, machine learning, neural networks, data mining, gene expressions, pattern or face recognition, cognitive psychology, and astronomy.

50. A method for classifying a subject in need thereof into a population based on a diagnostic system comprising:

(a) obtaining a sample for classifying;

(b) performing an assay on the sample;

(c) providing or generating a plurality of data points from two or more different populations with the assay, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a*x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the assay results from the sample; and

(i) determining the distance from the sample's assay results to each of the two or more populations, wherein the shortest distance from the sample's assay results to the population is the classification.

51 . The method according to claim 50, wherein the diagnostic system is chosen from Alzheimer's diagnostic assays, machine learning, neural networks, data mining, gene expressions, pattern or face recognition, cognitive psychology, and astronomy.

Description:
METHODS FOR CLASSIFYING POPULATIONS INCLUDING

ALZHEIMER'S DISEASE POPULATIONS

[0001 ] The present application claims the benefit of priority under 35 U.S.C. § 1 19 to U.S. Provisional Patent Application No. 62/129,715, filed March 6, 201 5, and U.S.

Provisional Patent Application No. 62/198,838, filed July 30, 201 5; the contents of each respective application is incorporated herein by reference.

[0002] BACKGROUND

[0003] Alzheimer's disease (AD) is a neurodegenerative disorder characterized by the progressive decline of memory and cognitive functions. It is estimated that over five million Americans are living with this progressive and fatal disease. Alzheimer's destroys brain cells, causing memory loss and problems with thinking and behavior that decrease quality of life. AD has no known cure, but treatments for symptoms can improve the quality of life of the millions of people, and their families, suffering from AD. An early diagnosis of AD gives the patient time to make choices that maximize quality of life, reduces anxiety about unknown problems, gives more time to plan for the future, and provides a better chance of benefiting from treatment.

[0004] The inaccuracy of the diagnosis for AD, however, makes therapeutic intervention difficult, particularly at early stages to prevent significant neurodegeneration and cognitive dysfunction. Thus, there is a need for highly sensitive and specific tests to diagnose Alzheimer's disease.

[0005] Biomarker assays for AD are affected by the input variables related to the status of the cells before using in the assay. F.V. Chirila et al., J. Alzheimer's Disease 33, 165-176 (2013); F.V. Chirila et al., J. Alzheimer's Disease 42, 1279- 94 (2014); W.Q. Zhao et al., Neurobiol. Dis. 1 1 , 166-183 (2002); T.K. Khan et al., Proc. Natl. Acad. Sci. U.S.A. 103(35), 13203-13207 (2006); T.K. Khan et al., Neurobiol Aging 2, {<o), 889-900 (2008); T. K. Khan et al., Neurobiol. Dis. 34(2), 332- 329 (2009). Such input variables include cell density (cell-cell interaction), age of the patients, passage number (number of cell population duplication), as well as other assay ingredients such as Matrigel™, Dulbecco's Modified Eagle Medium (DMEM), or Fetal Bovine Serum (FBS). Very often, a cell based assay will have to undergo long multivariate studies before reaching a practical value.

[0006] For example, FBS lot-to-lot variation often changes assay outputs and shifts cutoffs. Such shifting is shown in Figures 1 (A), (C), and Figures 2(A) - (B), where the dependence of the human skin fibroblast aggregation, measured by the natural logarithm of the unit aggregate area, Ln(Area/Number), on the Ln(Cell Density) is shown. F.V. Chirila et al., J Alzheimer's Disease 33, 165-176 (201 3); F.V. Chirila et al., J. Alzheimer's Disease 42, 1279-94 (2014). For this study, five FBS lots were used that originated from three different companies (i.e., Gemini Bio Products, Gibco Laboratories, and Atlanta Biologicals, Inc.).

[0007] However, e.g., in the case of diagnostic assays for Alzheimer's disease, it can be a long and laborious process to find an FBS lot that provides a good dynamic range (e.g., a good separation between AD and AC and/or non-ADD populations) for the output of the cell-based assays (e.g., Ln(A/N) for cell aggregation assays and fractal dimension for the network complexity assay). For example, 10 to 1 5 FBS lots will have to be tested and then ranked in terms of dynamic range, separability between different classes of patients, coefficient of variation, linearity etc. This approach translates into delays in reaching an assay's practical value. If the assay is already available commercially, then getting the results for patients will be delayed whenever the FBS lot is changed due to the functional quality control (QC) required for choosing another FBS lot. Furthermore, the lot-to-lot FBS variation might make impossible the use of a fixed cutoff in the case when the assay is used to discriminate between different classes of patients, and the cutoff will have to be readjusted each time a new lot is used.

[0008] One strategy to bypass the lot-to-lot variation is to replace the FBS with a serum free media. D.W. Jayme et al., Nature 334, 547-548 (1988); H. Haniu et al., Toxicol In Vitro 27(6), 1679-85 (201 3); E. Falkner et al., Toxicol In Vitro 20, 395-400 (2006); D. Brunner et al., AL TEX 27(1 ), 53-62 (2010); J. van der Valk et al., Toxicol In Vitro. 24(4), 1053-1063 (2010); D. W. Jayme et al., Cytotechnology 33(1 -3), 27-36 (2000). This strategy allows the cells to grow more uniformly. As a result, some of cell's properties will be less prone to changes in the FBS. However, this strategy is more labor intensive, more expensive, and it might not be successful for all the cell- based assays due to the controlled nature of the growth factors. From a practical point of view, more expensive and more labor-intensive methods translate into higher costs and lower profit margins. Furthermore, there are situations where serum is preferred to serum-free media. H. Haniu et al., Toxicol In Vitro 27(6), 1 679-85 (2013).

[0009] Another strategy, through the method disclosed herein, provides the same cutoff regardless of the FBS lot, making the comparison between FBS lots easier.

[0010] During the validation process of an AD diagnostic assay disclosed herein, the inventors surprisingly discovered that the classification of patients depends on two unexpected variables. The first variable, x, is the status of the cells before the experiment, measured by the natural logarithm of cell density. The second variable, y, is one of the ingredients of the media feeding the human skin fibroblast cells called fetal bovine serum (FBS). The dependence on the first variable, Ln(Cell density), is linear, while the second variable, FBS lot, alters the linear parameters of the data classes in a discreet manner (Figure 1 (A), (C) and Figures 2(A) - (B)).

[001 1 ] To determine the optimum diagnosis, i.e., the separation between two classes of patients when their linear dependence changes discreetly with a second variable, the inventors developed a two-stage method that results in an optimum choice with respect to the second variable.

[0012] Accordingly, the present inventors developed new methods for classifying two or more AD populations, such as AD class 1 (C1 ) and Age-matched control class 2 (C2), based on one or more biomarkers. For example, the present inventors surprisingly discovered a novel two-stage method that accounts for the dependence of the diagnostic assay measurements, Ln(Area/Number) (see F.V. Chirila et al., J. Alzheimer's Disease 33, 165-176 (2013); F.V. Chirila et al., J Alzheimer's Disease 42, 1279-94 (2014)), on cell density before the experiment and FBS. As it turns out, those new methods also can be applied to other fields of study such as machine learning, neural networks, data mining, gene expression, pattern or face recognition, cognitive psychology, or astronomy.

[0013] In some embodiments, disclosed herein is a method for classifying an Alzheimer's disease (AD) population from an age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing the biomarker assay on the biological samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

[0014] Also disclosed herein is a method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) obtaining a biological sample from the subject;

(b) performing a biomarker assay on the biological sample;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function; (f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

[0015] In various embodiments, disclosed herein is a method for classifying an Alzheimer's disease (AD) population from an age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, and Alzheimer's disease specific molecular biomarkers (ASDMB) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and (f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

[0016] In some embodiments, disclosed herein is a method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising

(a) obtaining a biological sample from the subject;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, and Alzheimer's disease specific molecular biomarkers (ASDMB) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function; (h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

[0017] In various embodiments, disclosed herein is a method of establishing a ranked order for two or more Fetal Bovine Serum (FBS) lots using an assay with FBS comprising:

(a) obtaining biological samples from two or more different populations;

(b) performing the assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the two or more different populations are separated; and

(g) plotting a dynamic range of the two or more different populations, wherein the plot results in a ranked order of the two or more FBS lots.

[0018] In some embodiments disclosed herein is a method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot; (b) performing the assay with the untested FBS lot;

(c) providing or generating a plurality of data points from two or more populations with the biological assay using two or more additional FBS lots, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the subjects of the two or more populations are separated;

(h) normalizing the biomarker result from the untested FBS lot;

(i) plotting a dynamic range of the two or more additional FBS lots and the untested FBS lot, wherein the plot results in a ranked order of the total FBS lots.

[0019] In various embodiments disclosed herein is a method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot;

(b) performing the biological assay with the untested FBS lot;

(c) normalizing the biomarker result from the untested FBS lot;

(d) plotting the untested FBS lot normalized result on a ranked order of tested FBS lots. [0020] In some embodiments, disclosed herein is a method for classifying two or more different populations based on a diagnostic system comprising:

(a) obtaining samples for classification from the two or more different populations;

(b) performing an assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input variable, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step (d) by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the method results in the separation of the two or more different populations.

[0021 ] In various embodiments disclosed herein is a method for classifying a subject in need thereof into a population based on a diagnostic system comprising:

(a) obtaining a sample for classifying;

(b) performing an assay on the sample;

(c) providing or generating a plurality of data points from two or more different populations with the assay, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable; (d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the assay results from the sample; and

(i) determining the distance from the sample's assay results to each of the two or more populations, wherein the shortest distance from the sample's assay results to the population is the classification.

[0022] Other aspects and embodiments of the present disclosure are set forth or will be readily apparent from the following detailed description. It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to be restrictive of the claims.

BRIEF DESCRIPTION OF THE FIGURES

[0023] Figure 1 (A) shows the dependence of the human skin fibroblast aggregation in an assay for AD skin samples compared to AC samples, as measured by the natural logarithm of the unit aggregate area, Ln(Area/Number), on the Ln(Cell Density); Figure 1 (B) illustrates a normalized assay (slope = 0); Figure 1 (C) shows the noise added to the slope and intercepts of fit lines from Figure 1 (A); Figure 1 (D) shows a normalized assay for the noisy data classes from Figure 1 (C).

[0024] Figures 2(A) - (B) show the dependence of Ln(Area/Number) on the Ln(Cell Density) and FBS lot for AD skin samples compared to AC samples; Figure 2(C) illustrates the probability distribution for the raw data from Figure 2(B); Figure 2(D) illustrates the probability distribution for the normalized data after the first stage of the method.

[0025] Figures 3(A) - (F) show a step-by-step description of the first stage of the analysis in the method for two noisy data classes of patients from Figure 1 (C).

[0026] Figure 4(A) shows the raw data ranked by the distance for the various FBS lots examined with the unknown; Figure 4(B) shows the average coefficient of variation normalized by the distance for the data Figure 4(A) including the unknown.

[0027] Figures 5(A) and (B) plot the CutOff(x) function for the FBS lot as a function of Ln(Cell Density).

[0028] Figure 6(A) shows the inverse slope versus the intercept for the fractal curves for AD samples, AC samples, and Non-Alzheimer's disease demented samples; Figure 6(B) shows normalized data from Figure 6(A); Figure 6(C) shows 120 randomly generated surrogate data; Figure 6(D) shows the normalized data from Figure 6(C).

[0029] Figure 7(A) ranks distances between three pairs of randomly generated data; Figure 7(B) shows the ranked distances have a linear dependence, D=D(Rank y ).

[0030] Figures 8(A) and (B) show the signal to noise ratio in establishing the discrimination limit (d-limit), where the level of noise is 10%.

[0031 ] Figure 9(A) shows Ln(Area/Number) versus the Ln(Cell Density) for AD samples (Ci), AC samples (C 3 ), and a third group labeled M; Figure 9(B) maps the three classes from Figure 9(A) into horizontal and parallel data classes; Figure 9(C) shows the dependence of total Length(L) on the slope, where the location of the three classes from Figure 9(A) are shown with a square(Ci), circle(M), and triangle(C 3 ); Figure 9(D) shows the dependence of the Length(L) on the X Projection/X range from Figure 9(A) and on the slope; Figure 9(E) shows the distance between pairs of segments from Figure 9(A) and Figure 9(B), where the solid black curves are the linear and exponential fits; Figure 9(F) shows the ratio of the distances in Figure 9(E).

[0032] Figure 10(a) shows raw data for two cell cycle-regulated genes of the Yeast Saccharomyces cerevisiae. See P.T. Spellman et al., Molec. Biol, of the Cell 9, 3273-3297 (1 998); Figure 10(b) shows rotation curves of low surface brightness galaxies. See K. de Naray et al., ApJS 1 65, 461 -479 (2006); K. de Naray et al., ApJS 676, 920-943 (2008); Figure 10(c) shows optimized data for the classes in Figure 10(A); Figure 1 0(d) shows optimized data for the classed in Figure 10(b).

[0033] Figures 1 1 (a) - (f) show the change in mini-mental state examination (MMSE) or Folstein test after three hours of administering Bryostatin or Placebo versus the change in Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) at 48 hours, where the dependence for both groups is linear.

[0034] Figures 12(a) and (b) plot the raw data and algorithm data for the method of classifying the age dependence for an Alzheimer's disease diagnostic assay.

DESCRIPTION

[0035] The following are definitions of terms used in the present specification. The initial definition provided for a group or term herein applies to that group or term throughout the present specification individually or as part of another group, unless otherwise indicated.

[0036] Abbreviations: AD: Alzheimer's disease; AC: non-demented age-matched control; Non-ADD: non-Alzheimer's dementia; S = any diagnostic system; Ci , C 2 , . . ., C n : two or more classes of data; D = distance between data classes. [0037] As used herein, the singular forms "a," "an," and "the" include plural reference unless the context dictates otherwise.

[0038] As used herein, Alzheimer's disease population can mean an Alzheimer's disease patient population, an age-matched control (AC) population, and/or a non- Alzheimer's disease demented (non-ADD) population.

[0039] As used herein, the term "subject" generally refers an organism. A subject can be a mammal or mammalian cell, including a human or human cell. The term also refers to an organism, which includes a cell or a donor or recipient of such cell. In various embodiments, the term "subject" refers to any animal (e.g., a mammal), including, but not limited to humans, mammals and non-mammals, such as non- human primates, mice, rabbits, sheep, dogs, cats, horses, cows, chickens, amphibians, and reptiles, which is to be the recipient of a compound or pharmaceutical composition described herein. Under some circumstances, the terms "subject" and "patient" are used interchangeably herein in reference to a human subject.

[0040] The present inventors discovered a new method for classifying an Alzheimer's disease (AD) population from an Age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, AC population, and/or the non-ADD population;

(b) performing the biomarker assay on the biological samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable; (c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

[0041 ] In some embodiments, for the first stage of the method, a novel approach of normalization based on distance preserving (i.e., isometric transformations) of the two or more classes of patients by the intercepts is provided. See H.S.M. Coxeter, Introduction to Geometry, John Wiley & Sons. Inc., New York (1961 ); I.M. Yaglom, Geometric Transformations I, II, I II, Mathematical Association of America). This first stage of the method is the normalization of the two or more data classes, i.e., slope = 0, and establishing of a fixed cutoff. However, because of the first stage, the separation of the two classes of patients increases exponentially. At the end of this stage, the data classes are parallel, horizontal, symmetrical with respect to the fixed cutoff, and classified by the intercepts. Therefore, in some embodiments, the first stage of the method uses four isometric transformations, normalizes across the slope, sets a fixed cutoff, and increases exponentially the distance between classes of patients.

[0042] In some embodiments, the first stage serves as a sorting procedure for the intercepts of the linear fits in a wide range of diagnostic systems (see Figures 3 and 10). In various embodiments, these diagnostic systems are chosen from: (1 ) AD diagnostic assays (F.V. Chirila et al., J. Alzheimer's Disease 33, 165-1 76 (2013); F.V. Chirila et al., J. Alzheimer's Disease 42, 1 279-94 (2014)); (2) machine learning (D. Elizondo, IEEE TRANSACTIONS ON NEURAL NETWORKS 17(2), 330-344 (2006)); (3) neural networks (F. Lavigne et al., Front Psychol. 5(842), 1 -24 (2014); D.L. Alkon et al., Biological Plausibility of Synaptic Associative Memory Models (1994) in Learning As Self-organization 247-259 (2013); D.L. Alkon et al., Biol. Cybern. 62, 363-376 (1990); M. Blair et al., Memory and Cognition 29(8) 1 153-1 164 (2001 )); (4) data mining (R. Haralick et al., Pattern Recognition 3587, 132-14

(2005) ); (5) gene expressions (S. Tavazoie et al., Nat. Genet. 22(3), 281 -285 (1999); P.T. Spellman et al., Molec. Biol, of the Cell 9, 3273-3297 (1998)); (6) pattern or face recognition (A.L. Yarbus et al., Biofizika. 6(2), 52-56 (1961 ); K.T. Blackwell et al., J. Experimental & Theoretical Artificial Intelligence 9(4), 491 -508 (1 997)); (7) cognitive psychology ((F. Lavigne et al., Front Psychol. 5(842), 1 -24 (2014); D.L. Alkon et al., Biological Plausibility of Synaptic Associative Memory Models (1994) in Learning As Self-organization 247-259 (2013); D.L. Alkon et al., Biol. Cybern. 62, 363-376 (1990); M. Blair et al., Memory and Cognition 29(8), 1 1 53-1 164 (2001 ); J. B. Talcott et al., Neuropsychologia 51 (3), 472-481 (2013)); and (8) astronomy (I.M. Yaglom, Geometric Transformations I, II, III, Mathematical Association of America; D. Elizondo, IEEE TRANSACTIONS ON NEURAL NETWORKS 17(2) 330-344

(2006) ; F. Lavigne et al., Front Psychol. 5(842), 1 -24 (2014); D.L. Alkon et al., Biological Plausibility of Synaptic Associative Memory Models (1994) in Learning As Self-organization (2013)). [0043] The present inventors further discovered a method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) obtaining a biological sample from the subject;

(b) performing a biomarker assay on the biological sample;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

[0044] For the second stage of the method, the normalized data classes with a fixed cutoff that resulted from the first stage are compared with respect to the second variable, e.g., fetal bovine serum. In some embodiments, this second stage employs two measures for classification: (1 ) the distance between data classes, D=[(X C i- Xc2) 2 +(Yci-Yc2) 2 ] V2 , and (2) the average coefficient of variation normalized by the distance between data classes, <CV>/D= [(CVci+CV C2 )/2]/D (Figures 1 (B) and (D)). Accordingly, in some embodiments, the second stage may use the distance between classes of patients, and the average coefficient of variation normalized by the distance, and rank them with the discreet variable, y, (e.g., fetal bovine serum lot). This dependence, called standard curve, D=D(Rank y ), is linear.

[0045] In general terms, the methods disclosed herein comprise ranking the states of a system, S, that separates two data classes, (C1 , C2), which show a linear dependence in one of the variables, x, and for which the linear dependence changes discreetly with a second variable, y. In some embodiments, the separation of the two classes of patients for the system, S=S(C1 , C2, x, y), measured by the distance between data classes, D, is ranked with respect to the discreet variable, y, D=D(Ranky).

[0046] In some embodiments, AD diagnostic assays disclosed herein establish a quantitative framework for diagnosing patients when dealing also with unexpected variables such as, cell density, FBS, age, etc. This strategy for AD diagnostic assays includes significantly reducing the quality control (QC) duration when changing the FBS lot. In some embodiments, the disclosed method requires a smaller number of samples needed in each class of patients, <5, for establishing the linear fits. The reduced QC duration increases the practical value of the assay, as it may help laboratories using cellular based assays and experiencing output shifts with FBS lots, or different varieties of FBS free media. In some embodiments, the method disclosed herein may be applied to AD diagnostic assays of the type , C 2 ,..., C n , x, y). [0047] The methods described herein have a general applicability to any system showing a linear dependence or a linear input-output function. For example, the method has no restriction in terms of the number of data classes that need to be separated. In some embodiments, the two-stage procedure may be employed in fields such as machine learning, neural networks, data mining, gene expression, pattern or face recognition, cognitive psychology, or astronomy. For example, in various embodiments, the two-stage analysis may be utilized by FBS and FBS free media vendors to provide a standardized method for correcting for the variation through isometric transformations. In some embodiments, a method for screening at least one FBS lot by employing the two-stage procedure is disclosed herein.

[0048] In some embodiments, disclosed herein is a method of establishing a ranked order for two or more Fetal Bovine Serum (FBS) lots using an assay with FBS comprising:

(a) obtaining biological samples from two or more different populations;

(b) performing the assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; (f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the two or more different populations are separated; and

(g) plotting a dynamic range of the two or more different populations, wherein the plot results in a ranked order of the two or more FBS lots.

[0049] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[0050] In some embodiments, the samples comprise human skin fibroblast cells.

[0051 ] In various embodiments, the dynamic range is based on the distance between the two or more different populations. In various embodiments, the dynamic range is based on the average coefficient of variation normalized by the distance.

[0052] Also disclosed herein is a method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot;

(b) performing the assay with the untested FBS lot;

(c) providing or generating a plurality of data points from two or more populations with the biological assay using two or more additional FBS lots, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept; (e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the subjects of the two or more populations are separated;

(h) normalizing the biomarker result from the untested FBS lot;

(i) plotting a dynamic range of the two or more additional FBS lots and the untested FBS lot, wherein the plot results in a ranked order of the total FBS lots.

[0053] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[0054] In some embodiments, the samples comprise human skin fibroblast cells.

[0055] In various embodiments, the dynamic range is based on the distance between the two or more different populations. In various embodiments, the dynamic range is based on the average coefficient of variation normalized by the distance.

[0056] In some embodiments, disclosed herein is a method of ranking at least one untested Fetal Bovine Serum (FBS) lot using an assay with FBS comprising:

(a) obtaining an untested FBS lot;

(b) performing the biological assay with the untested FBS lot;

(c) normalizing the biomarker result from the untested FBS lot; (d) plotting the untested FBS lot normalized result on a ranked order of tested FBS lots. In some embodiments, the assay is a diagnostic assay, such as a diagnostic assay for Alzheimer's disease.

[0057] In some embodiments, the isometric transformations in the method may be used for automation. The normalization of data classes across the slope and establishment of a fixed cutoff allows for comparison and classification with respect to the second variable, y (e.g., FBS lot). In some embodiments, the method disclosed herein further establishes a reference standard, i.e., a linear dependence between the dynamic range, D, and the rank of the discreet variable, y, D=D(Rank y ). Any other new condition in the discreet variable y, (e.g., FBS lot), may be compared with this reference standard. In some embodiments, the rank of an untested FBS lot can be determined immediately if the new y-state (e.g., FBS lot), has a good dynamic range, D, and a small level of noise by the normalized coefficient of variation, <CV>/D, based on the location on the linear standard curve (see Figure 4).

[0058] In some embodiments, the sample comprises at least one cell obtained from a human subject. In some embodiments, the at least one cell is a peripheral cell (i.e., a cell obtained from non-CNS tissue). In some embodiments, the at least one cell is a fibroblast cell. In various embodiments, the fibroblast cell is a skin fibroblast cell. Cell precursors of fibroblasts, such as induced pluripotent stem cells (IPSC) may also be used. For example, recent techniques for obtaining IPSC from human skin fibroblasts permitted differentiation of IPSC in cells such as neurons and showed imbalances in Αβ in both skin fibroblasts and IPSC differentiated neurons. Whalley K., "Neurodegenerative disease: Dishing up Alzheimer's disease," Nature Reviews Neuroscience 13, 149 (March 2012) | doi:10.1038/nrn3201 . [0059] In some embodiments, the sample comprises at least one cell chosen from skin cells, blood cells (lymphocytes), and buccal mucosal cells.

[0060] The non-Alzheimer's disease cells may be chosen, e.g., from an age- matched control. In some embodiments, the age-matched control is chosen from a non-AD non-demented population. In some embodiments, the age-matched control is chosen from a non-AD demented population, such as patients with Huntington disease or Parkinson's disease.

[0061 ] The at least one cell may be cultured in a media for growth. In some embodiments, the media comprises FBS. In some embodiments, the cells are cultured in media that is a protein mixture, such as a gelatinous protein mixture. A non-limiting exemplary gelatinous protein mixture is Matrigel™. Matrigel™ is the trade name for a gelatinous protein mixture secreted by the Engelbreth-Holm-Swarm (EHS) mouse sarcoma cells and marketed by BD Biosciences. This mixture resembles the complex extracellular environment found in many tissues and is used by cell biologists as a substrate for cell culture.

[0062] In some embodiments, the at least one cell is cultured in a preparation comprising extracellular matrix proteins. In some embodiments, the preparation comprises laminin, collagen, heparin sulfate proteoglycans, entactin/nidogen, and/or combinations thereof. In some embodiments, the preparation is extracted from a tumor, such as the EHS mouse sarcoma. The preparation may further comprise a growth factor, such as TGF-beta, epidermal growth factor, insulin-like growth factor, fibroblast growth factor, tissue plasminogen activator, and/or other growth factors or combinations thereof. In certain embodiments, the growth factors occur naturally in the EHS mouse sarcoma. Extracellular matrix proteins may also contain numerous other proteins. [0063] In some embodiments, the at least one cell is cultured in a basement membrane preparation. In some embodiments, the preparation is solubilized. In some embodiments, the basement membrane preparation is extracted from a tumor, such as the EHS mouse sarcoma— a tumor rich in extracellular matrix proteins. Its major component is laminin, collagen IV, heparin sulfate proteoglycans, and entactin/nidogen. In some embodiments, the preparation contains TGF-beta, epidermal growth factor, insulin-like growth factor, fibroblast growth factor, tissue plasminogen activator, and/or other growth factors which may or may not occur naturally in the EHS tumor. BD Matrigel Matrix Growth Factor Reduced (GFR) is found to be particularly well suited for applications requiring a more highly defined basement membrane preparation.

[0064] Within a short time after being cultured, measurable cellular networks form. This time period may vary in view of, for example, cell type and conditions, but generally, this time period ranges from about 1 hour or less, ranging from about 10 minutes to about 60 minutes, such as from about 10 minutes to about 45 minutes or any time in between. After a time, for example, approximately 5 hours, these networks start to degenerate and edges retract to leave behind measurable "clumps" or aggregates. In some embodiments, the time period for culturing the at least one cell is chosen from about 1 hour to about 72 hours, such as from about 12 hours to about 72 hours or from about 24 hours to about 48 hours. In various embodiments, the time period is about 48 hours or any one hour increment subdivision thereof.

[0065] The method(s) may further comprise imaging the cultured cells at the end of the time period. Images may be captured according to techniques known in the art. For example, images of the cellular networks may be captured with an inverted microscope, such as Western Digital AMID Model 2000, and controlled by a computer via image acquisition software at a desired magnification. Appropriate imaging techniques include, but are not limited to, confocal microscopy, phase contrast, bright field, fluorescence, differential interference contrast, and robotic systems.

[0066] In some embodiments, the methods disclosed herein classify two or more AD populations based on a biomarker assay. Biomarker assays can be any AD diagnostic assay. For example, AD diagnostic assays include, but are not limited to, cell aggregation, fractal dimension, Protein Kinase C (PKC) epsilon, Alzheimer's disease specific biomarkers (ADSMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining Initiative (ADNI) biomarkers.

[0067] Cell Aggregation

[0068] In some embodiments, the AD diagnostic assay is cell aggregation. For example, the area of aggregates can be determined by any suitable method, e.g., by fitting an ellipse across the aggregate. The counting of aggregates as well as aggregate area determination can be performed manually or can be automated, e.g., by image processing techniques known in the art.

[0069] In some embodiments, cell density is measured based on the number of cells per μιτι 2 or per field of view. In certain embodiments, cell density is measured by measuring the number of cells per 10χ image. In certain embodiments, the rate of change of the average area per number of aggregates as a function of cell density is evaluated within the boundaries of 320 to 550 cells/1 Ox image or such as, of 330 to 500 cells/1 Ox image.

[0070] Cell aggregation rate is determined by evaluating the rate of change of the average area per number of aggregates as a function of cell density. In some embodiments, the rate of change of the average area per number of aggregates as a function of cell density is evaluated by determining the slope of a linear fit between the average area per number of aggregates and cell density.

[0071 ] The aggregation rate of the cultured cells obtained from the human subject is compared to the aggregation rate determined using non-Alzheimer's disease control cells. The diagnosis is positive for Alzheimer's disease if the aggregation rate of the cultured cells from the human subject is increased compared to the aggregation rate determined using the non-Alzheimer's disease control cells.

[0072] Fractal Dimensions

[0073] In some embodiments, the AD diagnostic assay is fractal dimensions. For example, complexity of human skin fibroblast networks can be quantified by computing their fractal dimensions. Fractal analysis utilizes the complexity of the networks as means for distinguishing AD, AC, and non-ADD cells. Fibroblast cells obtained from patients suffering from AD have a statistically significant lower fractal dimension than AC cells when grown in tissue culture. The complexity of the networks measured by fractal dimension is also markedly different for fibroblasts taken from AD as compared to AC and non-ADD fibroblasts. Thus, a reduced complexity of human skin fibroblast networks AD cases provides distinctions from AC and non-ADD cases.

[0074] After network degeneration (e.g., about 48 hours), cells migrate and reach confluence within a few days. This recovery is captured by a linear increase in fractal dimension. Recovery as measured by the slope and intercept of the fractal curves therefore shows quantifiable differences between AD, non-ADD, and AC cells.

[0075] Fractal dimension can be generalized to N(S)=(1 /S)D, wherein D is the dimension and can be an integer or non-integer. Taking logarithms of both sides gives log(N(s))=D log( 1 /s), such that the fractal dimension can be determined by plotting log(N(s)) against log(1/s). If the slope is a non-integer, the dimension is a fractional (fractal) dimension.

[0076] The fractal dimension may be calculated using a standard box counting procedure after raw images (e.g., digital images) are filtered through an edge detection procedure that uses, for example, the difference of two Gaussians. Edge detection is a term used in the field of image processing, particularly in the areas of feature detection and feature extraction, to refer to algorithms which aim at identifying points in a digital image at which, for example, the image brightness changes sharply or has other discontinuities.

[0077] It can be shown that under rather general assumptions for an image formation model, discontinuities in image brightness are likely to correspond to one or more of discontinuities in depth, discontinuities in surface orientation, changes in material properties and variations in scene illumination.

[0078] Applying an edge detector to an image may lead to a set of connected curves that indicate the boundaries of objects, the boundaries of surface markings as well curves that correspond to discontinuities in surface orientation. Thus, applying an edge detector to an image may significantly reduce the amount of data to be processed, and may therefore filter out information that may be regarded as less relevant while preserving the important structural properties of an image. If the edge detection step is successful, the subsequent task of interpreting the information content in the original image may therefore be substantially simplified.

[0079] Methods for edge detection can generally be grouped into two categories: search-based and zero-crossing based. The search-based methods detect edges by first computing a measure of edge strength, usually a first-order derivative expression such as the gradient magnitude, and then searching for local directional maxima of the gradient magnitude using a computed estimate of the local orientation of the edge, usually the gradient direction. The zero-crossing based methods search for zero crossings in a second-order derivative expression computed from the image in order to find edges, usually the zero-crossings of the Laplacian or the zero crossings of a nonlinear differential expression. As a pre-processing step to edge detection, a smoothing stage, for example Gaussian smoothing, may be applied. In other embodiments noise filtering algorithms may be employed.

[0080] The edge detection methods that have been published mainly differ in the types of smoothing filters that are applied and the way the measures of edge strength are computed. As many edge detection methods rely on the computation of image gradients, they also differ in the types of filters used for computing gradient estimates in the x- and y-directions.

[0081 ] The fractal dimension is determined using a box counting procedure wherein the image is covered with boxes, for example, by a computer. The goal is to determine how the number of boxes needed to cover the image changes with the size of the boxes. If the object is 1 -dimensional, such as a line, the relationship is N(s)=(1/s) 1 as described above, and so on for higher dimensions. In some embodiments, the box counting procedure is implemented on a computer using digital images of cell samples.

[0082] A positive diagnosis for AD is made if the fractal dimension for Αβ-treated test cells is less than the fractal dimension for non-treated test cells. A positive diagnosis for AD is made if the difference in fractal dimension between Αβ-treated test cells and AD control cells is statistically significant.

[0083] Lacunarity [0084] In some embodiments, the AD diagnostic assay is lacunarity. As used herein, "lacunarity" refers to a measure of how a fractal fills space. It is used to further classify fractals and textures which, while they may share the same fractal dimension, appear very visually different. Dense fractals have a low lacunarity. As the coarseness of the fractal increases, the lacunarity decreases; intuitively from lacuna meaning "gap" (more gaps=higher lacunarity). See for example, U.S. Application Publication No. 2014/0248648, which is incorporated herein by reference.

[0085] Lacunarity is a complementary measure for complexity discrimination that quantifies the gaps in cellular networks. AD cell lines show an increased average lacunarity when compared with cell lines from AC and non-ADD individuals.

[0086] The lacunarity analysis method presently disclosed quantifies gaps of the fibroblast patterns as a complementary measure of complexity used as a second level of discrimination. The average lacunarity is higher for AD fibroblasts in comparison to AC and non-ADD fibroblasts. Typically the lacunarity increases and peaks when the network degeneration is maximized, i.e., only isolated aggregates are visible. Lacunarity drops as network regeneration starts.

[0087] A positive diagnosis for AD is made if the lacunarity for A. beta. -treated test cells is greater than the lacunarity for non-treated test cells. In some embodiments, a positive diagnosis for AD is made if the difference in lacunarity between Αβ treated test cells and AD control cells is statistically significant.

[0088] Cell Migration

[0089] In some embodiments, the AD diagnostics assay is cell aggregation. Cell migration allows for distinguishing between AD, AC, and non-ADD cells. A freely migrating cell is defined as a cell that is not attached to aggregates. Freely migrating cells may be counted at different times during culturing. For example, for times ti and t 2 , the migration rate is calculated as R=(N 2 -N 1 )/AT, where AT=t2-t1 is the time interval between cell migration counts and N1 and N2 are the number of cells migrating at the times t1 and t2. For example, in some embodiments, the number of freely migrating cells may be counted about 24 hours after plating, such as about 36 hours, about 48 hours, about 50 hours, about 52 hours, about 55 hours, about 57 hours, or about 60 hours after plating. In at least one embodiment, the migration rate is counted for times N 1 = 48 hours after plating and N 2 =55 hours after plating. In some embodiments, the initial cell density is controlled. In at least one embodiment, the initial cell density is controlled to about 50 cells/mm 3 .

[0090] A positive diagnosis for AD is made if the number of migrating cells is less than the number of migrating cells for age-matched controls (AC) or non-Alzheimer's disease demented (Non-ADD) patients. In some embodiments, a positive diagnosis for AD is made if the difference in number of migrating cells between AD and AC or between AD and Non-ADD cells is less than one standard deviation from the mean. In at least one embodiment, a migration rate lower than about 0.3 hr "1 is indicative of AD.

[0091 ] PKC Epsilon

[0092] In some embodiments, the AD diagnostic assay is PKC epsilon. For example, protein kinase C(PKC) isozymes particularly -a and -ε, play a critical role in regulating major aspects of AD pathology including the loss of synapses, the generation of Αβ and amyloid plaques, and the GSK^-mediated hyperphosphorylation of tau in neurofibrilliary tangles. Evidence of AD-specific signaling deficits has been previously found in peripheral tissues such as blood, skin fibroblasts, and ocular tissues. PKC-ε. is an accurate AD Biomarker in AD skin fibroblasts.

[0093] PKC epsilon levels are lower in Alzheimer's Disease subjects (AD) than in age matched controls (AC). That is, a method of diagnosing Alzheimer's Disease in a human subject comprises the steps of: a) determining the PKC epsilon level in said human subject; and b) comparing the PKC epsilon level in said human subject to the PKC epsilon level in a control subject; wherein said method is indicative of Alzheimer's Disease in said human subject if the PKC epsilon level in said human subject is lower than the PKC epsilon level in said control subject. See e.g., U.S. Patent Application Publication No. 2014/00381 86, the content of which is incorporated herein by reference.

[0094] Alzheimer's disease specific molecular biomarker (ADSMB)

[0095] In some embodiments, the AD diagnostic assay is Alzheimer's disease specific molecular biomarker (ADSMB) or AD-index. See e.g., U.S. Patent No. 7,595, 167 and U.S. Patent Application Publication No. 2014/0031245, which the contents for each is incorporated herein by reference. For example, diagnostic methods and methods of screening compounds useful for treating Alzheimer's disease are based on a unique molecular biomarker for Alzheimer's disease. The numerical value of the Alzheimer's disease-specific molecular biomarker (ADSMB) will depend on certain variables, such as, for example, the cells used in the assay, the protein kinase C activator used in the assay and the specific MAP kinase proteins that are targeted for measurement of phosphorylation ratios.

[0096] For example, the Alzheimer's disease-specific molecular biomarker may be measured by determining the ratio of phosphorylated Erk1 to phosphorylated Erk2 in cells that have been stimulated by Bradykinin, which is an inflammatory mediator, and subtracting from this the ratio of phosphorylated Erk1 to phosphorylated Erk2 in cells that have been stimulated with a control solution (vehicle) that lacks the Bradykinin (i.e., an inflammatory mediator). In certain embodiments, if the difference is greater than zero, i.e. a positive value, this is diagnostic of Alzheimer's disease. If the difference is less than or equal to zero, this is indicative of the absence of Alzheimer's disease.

[0097] The Alzheimer's disease-specific molecular biomarkers are measured by determining the ratio of two phosphorylated MAP kinase proteins after stimulation of cells with Bradykinin, which is an inflammatory mediator. The molecular biomarker may be measured by determining the ratio of a first phosphorylated MAP kinase protein to a phosphorylated second MAP kinase protein in cells that have been stimulated by Bradykinin (i.e., an inflammatory mediator) and subtracting from this the ratio of phosphorylated first MAP kinase protein to phosphorylated second MAP kinase protein in cells that have been stimulated with a control solution (vehicle) that lacks Bradykinin. In certain preferred embodiments, if the difference is greater than zero, i.e. a positive value, this is diagnostic of Alzheimer's disease. If the difference is less than or equal to zero, this is indicative of the absence of Alzheimer's disease.

[0098] The Alzheimer's disease-specific molecular biomarker is a positive numerical value in cell samples taken from patients diagnosed with Alzheimer's disease (AD cells). When the Alzheimer's disease-specific molecular biomarker is measured by determining ratios of phosphorylated Erk1 to phosphorylated Erk2 in AD cells that have been stimulated with bradykinin, the positive numerical values for the Alzheimer's disease-specific molecular biomarker in AD cells may range from about zero to about 0.5. [0099] The Alzheimer's disease-specific molecular biomarker is a negative numerical value when measured in cells taken from subjects diagnosed with non- Alzheimer's disease dementia (non-ADD cells), such as, for example, Parkinson's disease or Huntington's disease or Clinical Schizophrenia. When the Alzheimer's disease-specific molecular biomarker is measured by determining ratios of phosphorylated Erk1 to phosphorylated Erk2 in non-ADD cells that have been stimulated with bradykinin, the negative numerical values may range from about zero to about -0.2 or about -0.3.

[00100] The Alzheimer's disease-specific molecular biomarker may be a negative numerical value, zero or very low positive numerical value in cell samples from age- matched control cells (AC cells) taken from patients who do not have Alzheimer's disease. When the Alzheimer's disease-specific molecular biomarker is measured by determining ratios of phosphorylated Erk1 to phosphorylated Erk2 in AC cells that have been stimulated with bradykinin, the Alzheimer's disease-specific molecular biomarker in AC cells may range from less than about 0.05 to about -0.2.

[00101 ] The Alzheimer's disease-specific molecular biomarkers may be measured by calculating the ratio of phosphorylated Erk1 to phosphorylated Erk2 in cells that have been stimulated with bradykinin minus the ratio of phosphorylated Erk1 to phosphorylated Erk2 in cells that have stimulated with a solution lacking bradykinin. This is expressed as the following: Alzheimer's disease-specific molecular biomarker={(pErk1 /pErk2) bradyk i nin }-{(pErk1 /pErk2) veh icie}-

[00102] Protein kinase C activators that are specifically contemplated include, but are not limited to: Bradykinin; .alpha.-APP modulator; Bryostatin 1 ; Bryostatin 2; DHI; 1 ,2-Dioctanoyl-sn-glycerol; FTT; Gnidimacrin, Stellera chamaejasme L; (-)- Indolactam V; Lipoxin A.sub.4; Lyngbyatoxin A, Micromonospora sp.; Oleic acid; 1 - Oleoyl-2-acetyl-sn-glycerol; 4 .alpha.-Phorbol; Phorbol-1 2, 1 3-dibutyrate; Phorbol-12, 13-didecanoate; 4a-Phorbol-12, 13-didecanoate; Phorbol-12-myristate-13-acetate; L- a-Phosphatidylinositol-3, 4-bisphosphate, Dipalmitoyl-, Pentaammonium Salt; L- a- Phosphatidylinositol-4, 5-bisphosphate, Dipalmitoyl-, Pentaammonium Salt; L- a- Phosphatidylinositol-3, 4, 5-trisphosphate, Dipalmitoyl-, Heptaammonium Salt; 1 - Stearoyl-2-arachidonoyl-sn-glycerol; Thymeleatoxin, Thymelea hirsuta L.; insulin, phorbol esters, lysophosphatidylcholine, lipopolysaccharide, anthracycline dannorubicin and vanadyl sulfate. Also included are compounds known as "bryologues." Bryologues are described, for example, in Wender et al. Organic letters (United States) May 12, 2005, 7 (10) p1995-8; Wender et al. Organic letters (United States) Mar. 17, 2005, 7 (6) p1 1 77-80; Wender et al. Journal of Medicinal Chemistry (United States) Dec. 1 6, 2004, 47 (26) p6638-44. A protein kinase C activator may be used alone or in combination with any other protein kinase C activator in the diagnostic methods, kits and methods of screening compounds disclosed herein.

[00103] PKC Isozyme Index

[00104] In some embodiments, the AD diagnostic assay is PKC isozyme index. For example, the PKC isozyme index comprises measuring levels of steady state or phosphorylated PKC isozymes in peripheral cells from a candidate subject and, optionally, from a non-AD control subject (AC). See e.g., U.S. Patent Application Publication No. 201 1 /0212474, which is incorporated by reference herein. Sequentially or concurrently, steady levels of a first PKC isozyme are measured in peripheral cells from the AD and AC subjects both in the absence of, and in the presence of, an Αβ peptide to generate a first ratio of the PKC isozyme level (PKC isozyme level in the absence of Αβ peptide/level in the presence of Αβ peptide). A second PKC isozyme ratio is also obtained by measuring steady state or phosphorylated levels of a second PKC isozyme in peripheral cells from a subject, again in the absence of and in the presence of an Αβ peptide. Results of these measurements are then used to construct a third ratio, in which the first ratio (level of the first PKC isozyme obtained in cells not contacted with the Αβ peptide/level of the first PKC isozyme obtained in cells contacted with the Αβ peptide) is divided by the second ratio (level of the second PKC isozyme in cells not contacted with the Αβ peptide/level of the second PKC isozyme in cells contacted with the Αβ peptide) to generate a PKC Isozyme Index.

[00105] In some embodiments, the input variable is an independent variable that causes a change of the output or dependent variable (e.g., for an engine, the fuel-air ratio is an independent/input variable, while exhaust or power is the dependent/output variable). In an equation like (f(x)=a * x + b), for example, x is an independent/input variable for a linear function f, which is the dependent/output variable. In some embodiments, when diagnosing Alzheimer's disease, the input variable includes, but is not limited to, cell density (cell-cell interaction), age of the patients, passage number (number of cell population duplication), as well as other assay ingredients such as Matrigel™, Dulbecco's Modified Eagle Medium (DMEM), Fetal Bovine Serum (FBS), feeding time, or any variable in the cell cycle.

[00106] In some embodiments, the output variable is the dependent variable that changes as a result of the input/independent variable. For example, when diagnosing Alzheimer's disease the output variable includes, but is not limited to, the natural logarithm of (Area/Number) (e.g., when the biomarker is fibroblast aggregation and the input variable is the natural logarithm of cell density and fetal bovine serum), and the inverse slope (e.g., when the biomarker is fractal dimension, the input variable is the intercept for the fractal curves). [00107] In some embodiments, disclosed herein is method for classifying an Alzheimer's disease (AD) population from an Age-matched control (AC) population and/or a non-Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing the biomarker assay on the biological samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function,

wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

[00108] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[00109] In some embodiments, the samples comprise human skin fibroblast cells, which, may be cultured in fetal bovine serum.

[001 10] In some embodiments, the biomarker is chosen from cell aggregation, fractal dimension, protein kinase C epsilon, and Alzheimer's disease specific molecular biomarkers (ASDMB). In some embodiments, when the biomarker is fibroblast aggregation, the output variable is the natural logarithm of (Area/Number) and the input variable is the natural logarithm of cell density and fetal bovine serum. In various embodiments, when the biomarker is fractal dimension, the input variable is the intercept for the fractal curves and the output variable is the inverse slope.

[001 1 1 ] In some embodiments, the method further comprises establishing a universal cut-off value. In various embodiments, the performing step (c) comprises establishing a discrimination limit having a cut-off value based on the signal-to-noise ratio, and the plurality of data points generated from the AD population, AC population and/or the non-ADD population do not cross over the cut-off value.

[001 12] In various embodiments, disclosed herein is a method for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) obtaining a biological sample from the subject;

(b) performing a biomarker assay on the biological sample;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable; (d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

[001 13] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[001 14] In some embodiments, the samples comprise human skin fibroblast cells, which, may be cultured in fetal bovine serum.

[001 15] In some embodiments, the biomarker is chosen from cell aggregation, fractal dimension, protein kinase C epsilon, and Alzheimer's disease specific molecular biomarkers (ASDMB). In some embodiments, when the biomarker is fibroblast aggregation, the output variable is the natural logarithm of (Area/Number) and the input variable is the natural logarithm of cell density and fetal bovine serum. In various embodiments, when the biomarker is fractal dimension, the input variable is the intercept for the fractal curves and the output variable is the inverse slope.

[001 16] In some embodiments, the method further comprises establishing a universal cut-off value. In various embodiments, the performing step (c) comprises establishing a discrimination limit having a cut-off value based on the signal-to-noise ratio, and the plurality of data points generated from the AD population, AC population and/or the non-ADD population do not cross over the cut-off value.

[001 17] Also disclosed herein is a method for classifying an Alzheimer's disease (AD) population from an age-matched control (AC) population and/or a non- Alzheimer's disease demented (non-ADD) population based on a biomarker assay comprising:

(a) obtaining biological samples for classification from the AD population, the AC population, and/or the non-ADD population;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative (ADNI) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-inout, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function; (e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the method results in the separation of the AD population from the AC population and/or the non-ADD population.

[001 18] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[001 19] Also disclosed herein is a for classifying a subject in need thereof into an Alzheimer's disease (AD) population based on a biomarker comprising:

(a) providing skin fibroblast cells from the subject;

(b) performing a biomarker assay chosen from cell aggregation, fractal dimension, protein kinase C epsilon, Alzheimer's disease specific molecular biomarkers (ASDMB), lacunarity, cell migration, PKC isozyme index, and Alzheimer's disease Neuroimagining initiative (ADNI) on the cells cultured with Fetal Bovine Serum to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) providing or generating a plurality of data points from two or more AD populations with the biomarker assay, wherein the assay comprises an input variable and an output variable and wherein the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept;

(e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the biomarker results from the subject; and

(i) determining the distance from the subject's biomarker results to each of the two or more populations, wherein the shortest distance from the subject's biomarker result to the population is the diagnosis.

[00120] In some embodiments, a first plurality of data points comprises controls and a second plurality of data points comprises AD samples. In various embodiments, a first plurality of data points comprises controls, a second plurality of data points comprises AD samples, and a third plurality of data points comprises non-ADD samples. The non-ADD samples may be chosen from Huntington disease samples and Parkinson's disease samples.

[00121 ] In various embodiments, disclosed herein is a method for classifying two or more different populations based on a diagnostic system comprising:

(a) obtaining samples for classification from the two or more different populations; (b) performing an assay on the samples to generate a plurality of data points, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(c) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input variable, and "b" is the y-intercept;

(d) translating each plurality of data points to the origin by subtracting the intercept "b" for each respective linear function;

(e) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function; and

(f) reversing the downward translation step (d) by translating each plurality of data points upward by the intercept b for each respective linear function, wherein the method results in the separation of the two or more different populations. In some embodiments, the diagnostic system is chosen from Alzheimer's diagnostic assays, machine learning, neural networks, data mining, gene expressions, pattern or face recognition, cognitive psychology, and astronomy.

[00122] Also disclosed herein is a method for classifying a subject in need thereof into a population based on a diagnostic system comprising:

(a) obtaining a sample for classifying;

(b) performing an assay on the sample;

(c) providing or generating a plurality of data points from two or more different populations with the assay, wherein the assay comprises an input variable and an output variable and the assay output variable depends linearly on the assay input variable;

(d) fitting each plurality of data points with a linear function, f(x)=a * x + b, wherein "a" is the slope, "x" is the x-input, and "b" is the y-intercept; (e) translating each plurality of data points to the origin by subtracting the intercept b for each respective linear function;

(f) rotating each plurality of data points counterclockwise by-atan(a) for each respective linear function;

(g) reversing the downward translation step by translating each plurality of data points upward by the intercept b for each respective linear function;

(h) normalizing the assay results from the sample; and

(i) determining the distance from the sample's assay results to each of the two or more populations, wherein the shortest distance from the sample's assay results to the population is the classification. In some embodiments, the diagnostic system is chosen from Alzheimer's diagnostic assays, machine learning, neural networks, data mining, gene expressions, pattern or face recognition, cognitive psychology, and astronomy.

EXAMPLES

[00123] Materials and Methods

[00124] Banked and Fresh Cell Lines Used in This Study. Example experiments were carried out using skin fibroblasts samples from 40 patients - 24 banked samples as provided by the Coriell Institute for Medical Research (Camden, NJ) (Table 1 ), and 16 cases (Table 2) from the clinic provided by Marshall University (Huntington, WV).

[00125] The fibroblast cells were plated on a thick layer (-1 .8 mm) of 3-D matrix (Matrigel, BD Biosciences, San Jose, CA) on 1 2 well plates. See Chirila F.V. et al., J. Alzheimer's Disease, 33, 165-176 (201 3). The available patient information is posted on Coriell web site (http://ccr.coriell.org/). The cell lines analyzed (40) were based on a number of criteria including autopsy and genetic family history. A significant number of samples (7) were studied under double-blind conditions (Table 1 ), and a further sample confirmed the diagnostic differentiation on freshly obtained skin samples. The age-matched control (AC) samples were not demented at the date of skin biopsy extraction. All the samples were taken ante-mortem with one exception, AG08245 which was taken postmortem. The banked skin fibroblast cells were frozen stocks under liquid nitrogen. Primary cultures were established after thawing those frozen samples and followed through successive passaging. All cell lines used in this study were primary cell lines and were not treated in order to be immortalized.

[00126] Freshly taken fibroblasts were obtained as follows. Punch-biopsies (2-3 mm, upper arm) skin tissues from patients and controls were obtained. Cells with passages between 5 and 15 were used (Table 2).

[00127] The initial cell density was controlled to be 50 cells/mm 3 , and was homogenized with 1 .5 ml Dulbecco's Modified Eagle Medium with 10% fetal bovine serum and 1 % penicillin/streptomycin (PS) for each well. Cells were kept in a C0 2 water-jacket incubator (Forma Scientific) up to 7 days after plating.

[00128] Image capture. Images of the cellular networks were captured with an inverted microscope (Westover Digital AMID Model 2000, Westover Scientific, Bothell WA), controlled by a computer via an image acquisition software (Micron 2.0.0), using a 10x and a 4x objective. Five to nine images captured per well and three wells per cell line were typically used. In the first day, images were acquired every hour, the second day at every other hour, and the remaining three days at three times a day. Images were processed with ImageJ, freely available software from NIH (http://rsbweb.nih.gov/ij/).

[00129] Image quantification. The 5 images per well were initially taken using the same standard pattern, center (1 ), up (2), down (3), left (4), right (5) by moving one image field with respect to the central image. See F.V. Chirila et al., J. Alzheimer's Disease 33, 165-176 (2013). Later in the process, the number of images from 5 to 9 by filling the corners of the rectangle with images from 6 to 9, in order to increase the area investigated and further improve the coefficient of variation without affecting the diagnostic discriminability. Image 1 was always in the center of the well. To determine the center of the well, one of the following methods was used: a) the live image under 4x magnification should be symmetric, i.e. the shadows in the four corners should have equal areas for an aligned microscope; b) mark the center with a needle; c) use gridded plates (Pioneer scientific; Shrewsbury, MA) where the central square is always the 6th, in the central row or column. For image quantification, two sets of tools: initially manual as provided by Micron, software which came with the microscope; later automated with ImageJ, a freely available software from NIH (http://rsbweb.nih.gov/ij/).

[00130] For the initial cell count, a custom ImageJ plug-in was used in which "despeckle" was run three times; the image was filtered three times with a minimum filter of radius 0.5; and "Subtract Background" was run with a rolling radius of 20. Finally, the image was made binary and "Analyze Particles" was run in the size range 1 80-lnf inity. All of these ImageJ commands were run inside a loop to permit analysis of all the images from one cell line automatically. The ImageJ plug-in was tuned by using manual cell counts on the same images and the relative error was below 7%.

[00131 ] The target number of cells per 10x image was 41 7 which corresponded to an initial cell concentration of 50 cells/μΙ. A variation of cell concentration between 45 and 60 cells/μΙ was permitted. To minimize heterogeneity of the cell distribution in the image, images outside of the range 195-650 cells per 1 0x image were eliminated. For cellular aggregates at 48 hours, manual ellipse fitting with the Micron software was used.

[00132] Fractal dimension and lacunarity. For fractal and lacunarity analyzes FracLac_2.5 plug-in (http://rsbweb.nih.gov/ij/plugins/fraclac/fraclac.html) using the "box counting" method was used. The recovery slope and intercept were monitored by fitting a line in the range 20-80% of the min-max difference. The average lacunarity was calculated between 0 and 120 hours.

[00133] Average area per number of aggregates(A/N). Average area per number of aggregates was calculated in the following manner. For each image, an average aggregate area <A>, and a number of aggregates N, was calculated. Then, for each image, the ratio <A>, /N, was evaluated. Typically, five images per well were

_i ·Α ·. used and an average area per number for each well was evaluated as the 2. ■

3 - 4 - \

An additional average was performed over the three wells: ∑ — r I . Then the

natural logarithm of this measure, , was calculated for each cell line (Figure 1 ). The aggregates were manually fitted with ellipses using Micron 2.0 software and their area and number were recorded. An automatic script for ImageJ was developed which agreed well with manual ellipse fitting. These two approaches were within one standard deviation of each other.

[00134] Cell density. During the validation process, the inventors surprisingly discovered that the cell density is a hidden variable that can influence the output of the assay. This cell density is in T25 flasks before using the cells in the assay. The cell density was calculated based on the average cell number measured with the hemocytometer. The total number of cells was first estimated based on the volume of the medium used after trypsinization, then this total number of cells was divided by the surface of the flask 25 cm 2 . The cell density is also an indirect measure of the cell confluence.

[00135] Statistical Analysis. Separability between data classes is defined as the difference between the average fit functions in the range of interest (x m i n , x ma x), i.e., Sep={[f(Xmin)+ f(Xmin)/2 - [g(x m in)+ g(x m in)/2}, The overlap probability is calculated from a simple t-test function for 2-tailed and two samples with unequal variance. The details about the construction of the probability surface from Figure 2 are described in F.V. Chirila et al., J Alzheimer's Disease 33, 165-176 (201 3) and F.V. Chirila et al., J Alzheimer's Disease 42, 1279-94 (2014).

[00136] Data analysis. For data analysis, Gnuplot 4.4 was used. Gnuplot 4.4 is a freely available software (http://www.gnuplot.info). For fitting of the raw data points, a built in fit function from Gnuplot was used, which uses an implementation of the nonlinear least-squares (NLLS) Marquardt-Levenberg algorithm. Unless otherwise specified, the error-bars are standard errors of the mean (SEM). Gnuplot was chosen because every step can be visualized and saved as a source code in ".gnu" file format, making the recovery and visualization easy. However, this method can be easily implemented in other software of choice such as C, or C++. All the steps are standard because they depend on the slopes and intercepts of the data classes therefore prone to automation.

[00137] Probability distributions. For all the Alzheimer's disease patients (AD) and all the aged-matched-control patients (AC), the values for Ln(Cell Density) and Ln(Area/Number) were binned into intervals which are inversely proportional to the density of points, fit with Gaussian functions for each variable, and then integrated into a normalized two-dimensional distribution. [00138] Isometric transformations. The four Isometric transformation for the first stage of the algorithm were implemented step-by-step in Gnuplot and saved as gnu files for easy verification. These four isometric transformations are based on the initial fit parameters, slope and intercept, for each data class.

Objectives and Experiments

[00139] Objectives

[00140] The objectives of the study disclosed herein include validation for the peripheral biomarkers for AD using skin fibroblasts. The detailed experimental methods are briefly described here, and in detail in F.V. Chirila et al., J. Alzheimer's Disease 33, 165-176 (2013); F.V. Chirila et al., J. Alzheimer's Disease 42, 1 279-94 (2014). A goal of the validation study was to find the variables (e.g., hidden or unexpected variables) that may influence the diagnosis of AD, and then to quantify their dependence. Another goal was to apply corrective procedures that may remove an assay's dependence on those variables.

[00141 ] Also disclosed herein is the data analysis of some experiments performed during the assay validation that involved different FBS lots. Moreover, randomly generated data sets were exemplified to emphasize the generality of the methods disclosed herein, and the same linear ranking relationship, D=D(Rank y ), was discovered.

[00142] Examples

Example 1 : Alzheimer's disease diagnostic assay for different FBS lots

[00143] Figure 1 (A) and Figures 2(A) - (B) show the input-output function for an AD diagnostic assay for different FBS lots. In Figure 1 , the FBS A14 is from Gemini Bio Products (Triangles), the FBS 941 from Gibco Laboratories (Squares), and the FBS D1 1 from Atlanta Biologicals (Circles). In Figure 2, the FBS lot A92 is from Gemini Bio Products (Figure 2(A)), and the FBS lot 692 is from Gibco Laboratories (Figure 2(B)). In Figure 2, moreover, n represents the number of patients. The assay output, Ln(A/N), depends linearly on assay input, Ln(Cell Density), for the two classes of patients, class 1 -(AD; empty symbols) and class 2-(AC; filled symbols), and for the five FBS lots. As shown in Figure 1 (A) and Figures 2(A) - (B), the dependence of the assay output on the assay input is linear regardless of the FBS lot used. However, with each FBS lot there is a slight difference in the linear dependence, i.e., a difference in the slope and the intercept for each class of patients. This dependence on the FBS lot has practical implications because of the change in the cut-off, and therefore, in the diagnostic assay.

Example 2A: The first stage of the method

[00144] The first stage in a method disclosed herein comprises five steps {see, e.g., Figure 3).

[00145] Step 1 includes finding the representation in which the dependence/input- output function is linear and fitting the classes of patients with linear functions. For the two classes of patients, this representation is in a Ln(A/N) versus Ln(Cell Density) plot (see Figures 1 (A) and Figures 2(A) - (B)). To emphasize the generality of the method, three pairs of data classes were generated based on the fit lines of the raw data from Figure 1 (A). Then, noise was added to the slopes and intercepts (see Figure 1 (C)). Out of the three different conditions studied, one condition was picked (see the two lines corresponding to FBS lot 941 in Figure 1 (C)) to show as an example of the first stage of the method. The fit functions are f(x)=a * x+b for class 1 - AD skin samples (squares), and g(x)=c * x+d for the class 2-AC samples (circles) (Figure 3(A)). [00146] Step 2 includes translating the data classes to the origin i.e., by subtracting the intercepts i.e., "-b" and "-d." (See Figure 3(B)).

[00147] Step 3 includes rotating the data classes by the minus angles of the fit lines, i. e. "-atan(a)," and "-atan(c)" (see Figure 3(C)).

[00148] Step 4 includes reversing the shifting of the data with the intercepts from step 2 by translating the data by the intercepts "+b" and "+d" (see Figure 3(D)).

[00149] Step 5 includes-setting up the fixed cut-off in the middle of the gap "(b+d)/2," therefore translating the data by the difference [8-(b+d)/2] (see Figure 3(E)). The final step is depicted in Figure 3(F) and can be compared with the raw data classes from Figure 3(A).

[00150] Therefore, the first stage comprises a linearization and fitting step and four isometric transformations: translation, rotation, translation, translation.

[00151 ] The first four steps depend on the fit parameters i.e., slopes and intercepts, that make up the linearization and fitting steps. The last step in Figure 3 (step 5), has and arbitrary value for the Fixed Cutoff, which was assigned a value of 8. However, as long as the Fixed Cutoff is fixed for all of the different conditions in the second variable, y, herein the different FBS lots, the ranking and classification with respect to the second variable will not depend on the choice of the Fixed Cutoff. Example 2B: Results of the first stage of the method

[00152] In some embodiments, the first stage of the method produces: (a) a classification of the data sets by their intercepts i.e., "b" and "d"; (b) a slope normalization, slope = 0, and makes the two data classes parallel; (c) equidistant data classes with respect to the fixed cutoff, i.e., both data classes are at half difference between the intercepts, ((b-d)/2); and (d) a fixed cutoff (Figures 1 (B), (D)). The parameters used for the first three isometric transformations are set by the slopes and intercepts of the linear fits for each data class. The fourth isometric transformation has an arbitrary, but fixed cutoff. The classification with respect to this second variable does not depend on the arbitrary cutoff. Due to the standard nature of this first stage of the method, the result is a measurable distance, D, between the classes of patients (Figures 1 (B), (D)). In some embodiments, the output of the assay may be sorted as a function of the second variable, y, which may be, e.g., the FBS lot used (Figure 4).

Example 3A: Description of the second stage in the method

[00153] With a fixed cutoff and the two data classes parallel, horizontal (slope = 0), and equidistant, the classification with respect to the second variable, y (e.g., the FBS lot), may be performed (Figures 1 (B), (D)). The measure for the dynamic range between the two data classes of patients is the distance, D=[(X A D-XAC) 2 +(YAD- YAC) 2 ] V2 (Figure 1 (D), Figures 4(A) - (B)). For a measure of the assay noise level, the proposed measure is the average coefficient of variation for the two data classes normalized by the distance between data classes, <CV>/D= [100*(CV A D+CV A C)/2]/D, (Figure 1 (D), Figures 4(C) - (D)).

Example 3B: Results of the second stage in the method

[00154] Figures 4(A) and (C) show the raw data ranked by the distance, D, and the average coefficient of variation normalized by the distance, <CV>/D. Figures 4(B) and (D) show the same measures, the distance, D, and the average coefficient of variation normalized by the distance, <CV>/D, for the noisy data from Figure 1 (C). The linearity of both measures, the distance, D, and the average normalized coefficient of variation, <CV>/D, with respect to the second variable, y/FBS lot, is remarkable. Surprisingly, the noisy data from Figure 1 (C) yields the same ranking as the raw data, and a very close linear dependence allowing a clear classification with respect with the second variable (e.g., FBS lot). Furthermore, this linear dependence, D = D(Rank y ), shows that the optimum separation between classes of patients occurs, e.g., for low ranked FBS lots.

Example 4: Classification of an unknown y-state (e.g., FBS lot)

[00155] The distance between data classes, D, in their normalized form, after the first stage of the method depends linearly on the rank of second variable y (e.g., FBS lot). See, e.g., Figures 4(A) - (B).

[00156] Any new and untested FBS lot, y u , will fall on the same linear standard curve, D = D(Rank y ). The optimum FBS lots have a large dynamic range, D, i.e., a smaller rank, while sub-optimum lots have a smaller dynamic range, D, i.e., a bigger rank. This is called linear dependence, D = D(Rank y ), a reference standard. Regardless of the rank of the FBS lot, this representation can be used in the range where the signal plus noise does not cross over the cutoff line.

[00157] Additionally, an untested FBS lot can be classified by using the reference standard curve from Figure 4. For any untested FBS lot, approximately five samples/patients per data class are needed to determine the linear dependence for each class Ci ; f(x)=a * x+b, C 2 ; g(x)=c * x+d. In some embodiments, a range of 3 to 5 samples/patients per data class can be used in order to determine linear dependence. Once this dependence is established, the two-stage method described herein may be used to rank the new FBS lot, y u , on the reference standard curve, D=D(Rank y ) (see Figure 4, lines are best fits). The optimum y-state/FBS lot, has a good dynamic range, large D, and low noise, small <CV>/D, i.e., low Rank. Once the FBS lot is ranked accordingly (see, e.g., Figures 4(A) - (B) show the unknown FBS lot (U) with the arrow both plotted against D and the coefficient of variation), the separation boundary/cutoff function and the discrimination limit can be determined. Then, any unknown sample can be classified.

Example 5: Establishing the data class separation boundary-Cutoff function

[00158] From the practical and validation point of view, establishing a unequivocal function for defining the separation boundary/cutoff between two linear data classes is desirable {see, e.g., Thick lines in Figure 5). Given the linear dependence of the two classes of patients, on the first variable, x, the cutoff is assumed to also be a linear function, and may be found analytically as the location where the distances to the two classes of patients, i.e., fit lines, are equal, dAD = dAc- The equation for the cutoff is given by the equation for the angle bisector. Only the positive solution is usable for the cutoff locus in the method disclosed herein: CutOff (xH {c+a[(c?+ 1)/(c?+ 1)] 1/2 }x+d+b[(c 2 + 1)/(a 2 + 1)] 1/2 (Thick lines Figure 5).

[00159] In some embodiments, variants of this approach may be used by defining an area near the CutOff(x) line as a band of a certain percent formed of parallel lines shifted upward and downward with a percentage of the intercept, inside which the classes will be undetermined (see Dashed lines Figure 5(B)). However, this strategy depends on the dynamic range, D, of the two data classes needed to be compared, on the signal-to-noise ratio (SNR), <CV>/D, and on how close to the intersection point of the fit lines the two data classes are (see Example 10).

Example 6: Classification of an unknown sample

[00160] Once the linear dependence for the two classes of patients on the input variable is established, any unknown sample may be categorized as class 1 -AD or class 2-AC based on the shortest distance from its location, for example (x u , y u ), to the two fit lines, f(x)=a * x+b, g(x)=c * x+d (see Dotted/dashed lines in Figure 5(A)). The distance between the unknown sample located at (x u , y u ), and the fit line for the AC class is given by the perpendicular segment line to the AC fit line (see Dotted lines in Figure 5(A)). Similarly the distance from the unknown sample, (x u , y u ), and the fit line for the AD class is given by the perpendicular segment line to the AD fit line (see Dashed lines in Figure 5(A)). If the shortest distance is dAc, i.e., dAc < dAD, then the case is a class 2-AC (Circle in Figure 5(A)). If the shortest distance is dAD, i.e., dAD<dAc, then the case is a class1 -AD (Square in Figure 5(A)). The distances described herein, dAc and dAD, have an analytical description. The classification of an unknown sample can be up to the discrimination limit (see Example 10).

[00161 ] Examples of the concepts discussed above are illustrated in Figures 6(C) and (D), where the 120 samples randomly generated were classified as class 1 -AD, class 2-AC, class 3-Non-ADD (see F.V. Chirila et al., J. Alzheimer's Disease 33, 165-176 (2013)), on the basis of the cutoff functions between pairs off classes. The first stage of the procedure was to untangle and normalize the three data classes to further classify the assay performance, with other hidden variables, such as y/FBS lot.

Example 7: Establishing of a reference standard, D=D(Rank y )

[00162] In this example, the five FBS lots from three different companies are regarded as discreet values for the second variable, y"5¾{yi , y 2 , y3, y 4 , ys}. Figures 1 (A) and (C) illustrate the FBS lot variability for the three FBS lots. The unbiased classification with respect to this second variable is a result of this study (see Figure 4). In some embodiments, the first stage of the method allows for standardization by establishing a fixed cutoff (e.g., 8), which permits a fair and quantitative comparison of different FBS lots. In some embodiments, the second stage of the method allows the five FBS lots to be ranked (see Figure 4) based on the distance between data classes, D, and based on the normalized coefficient of variation, <CV>/D. Interestingly, the ranking shows a linear dependence with respect to the second variable, which is called a reference standard, D = D(Rank y ). Furthermore, this linear dependence is also followed for three pairs of randomly generated data sets (See Figure 7).

[00163] From the practical point of view, the method disclosed herein offers an advantage by reducing the QC time. Practically, the FBS lot-to-lot variation issue is removed. Moreover, any lot within the range described herein can be used after establishing the linear dependence within the data classes with a small number of samples, e.g., typically less than 5 samples per data class. In some embodiments, the signal to noise ratio may limit the operation range up to discrimination limit (see Example 10).

Example 8: Generalization of the first stage for three data classes (Fractal Dimension)

[00164] In some embodiments, the first stage of the analysis of the method was validated with a different AD diagnostic assay that quantifies the skin fibroblast network complexity with fractal dimension for different time points (F.V. Chirila et al., J Alzheimer's Disease 33, 1 65-176 (2013)). The fractal curves were constructed, and then the linear part of the curve was fitted with a line from which the slope and intercepts were extracted. Three classes of patients were studied with this assay, class 1 -AD, class 2-age-matched non-demented control (AC), and class 3-non Alzheimer's disease demented (Non-ADD), which includes Huntington disease (HD) and Parkinson's disease (PD) patients (see Figures 6(A) - (B)). For this assay, the linear representation is presented as 1 /slope versus the intercept of the linear part of the fractal curves. See, e.g., Figure 6(A) (showing raw data for AD samples (squares), AC samples (circles), and Non-ADD samples (triangles)). As shown in Figure 6(B), when using the first stage of the method for the three classes AD, AC, Non-ADD, the normalization and setting of a fixed cutoff are achieved. Furthermore, 120 randomly surrogate data were generated in this plane 1 /slope versus intercept (Figure 6(C)). These surrogate data play the role of unknown samples that may be tested with this assay. The cutoffs were established by using the equation of the angle bisector, then the first stage of the analysis of the method was used to normalize the data, and to set a fixed cutoff between two important data classes, AD and Non-ADD. See Figure 6(D). In some embodiments, this generalization sets the stage for applying the second part of method in a system where there are unexpected variables changing the linear dependence described in Figure 6(A). In some embodiments, the FBS lot is the unexpected variable in this assay. However, other variables such as protein concentration in Matrigel™ may also change the linear dependencies from Figure 6(A).

Example 9: Generality of the method

[00165] In some embodiments, the method disclosed herein is general and has applicability to other diagnostic systems, S, with 2 or more classes of patients, (Ci , C 2 ,..., Cn ) to be separated, and which show a linear dependence of the assay output on one of the variables, x, with discreet changes with a second variable, y"5¾{yi , y 2 , y3, y4, ys}- Furthermore, in some embodiments, the first stage of the method has no restrictions in terms of the number of data classes that need to be normalized. For the second stage of the method, in some embodiments, if more than two data classes need to be separated, then a decision must be made about the two most important classes that need to be separated. Figure 7 illustrates the first and second stages of the method for randomly generated data classes C1 and C2. To demonstrate the generality of the two stage procedure, two randomly generated data classes, as represented by circles, squares, and triangles, were considered from exponential functions of the type a * exp(-x/b), where a and b are parameters (see Figure 7(A); empty symbols represent the upper data sets and filled symbols represent the lower data sets). Class one was generated by the function c1 (x)={exp[- x/(2+0.2 * rand(0)]} + 2 * rand(0) * {exp[-x/(2+.2 * rand(0))} and class two was generated by the function c2(x)={0.2 * exp[-x/(3 + 0.3 * rand(0))]} + 2 * rand(0) * {0.2 * exp[-x/(3 + 0.3 * rand(0))]}, were rand(0) generates a random number between 0 and 1 . These randomly generated data classes show a linear dependence (Figure 7(A)). To mimic the change of the linear dependencies with the second hidden variable, y, three pairs of data classes labeled as red, green, and blue were generated (Figure 7(A)). After applying the two-stage procedure the similar linear dependence between the dynamic range, D, and the rank was determined, Figure 7(B), D=D(Rank y ). In some embodiments, the reference standard found for the AD diagnostic assay may be found for other systems requiring data classification and which have a change in the cutoff with hidden variables, linear in x, and changing the linear fit parameters with the second variable y.

[00166] An additional example is given in Figure 6 for three classes of data, out of which the separation between two groups, AD and Non-ADD patients, were identified as the two populations of interest. Based on this decision, the fixed cutoff is established between these two classes enabling a further quantification and ranking with any other hidden variable, y.

Example 10: The noise level establishes the discrimination limit/d-limit

[00167] In some embodiments, the level of noise with respect to the signal in the data may become important near the intersection point of the two linear classes of patients (see Figures 8(A) - (B)) where the two classes cross over the separation boundary/cutoff. The dashed vertical line in Figure 8(A) defines the discrimination limit (d-limit). To the right of the discrimination limit, the signal-to-noise ratio (SNR) is poor. For example, as can be seen, 1 1 out of the 51 data points from class 1 , and 18 out of 51 data points from class 2, are crossing over the cutoff line. The vertical dashed line from Figure 8(A) is located where the signal plus the noise intersects the cutoff line. The SNR ratio is closely related with the average coefficient of variation normalized by the distance, <CV>/D. Here, the d-limit suggests that high cell densities are not desirable for this diagnostic assay based on cell aggregation.

Example 11 : Theoretical considerations

[00168] The two fit lines for the FBS 941 from Figures 1 (A) and (C) corresponding to the two classes of patients, AD patients (C1 ; f(x)=ax+b), and AC controls (C3; g(x)=cx+d) were used as well as a line with an intermediate slope which was labeled (C2; l(x)=mx+n). These three data classes were used as raw data for the theoretical study disclosed herein. See Figure 9(A). The thickness of these fit lines were also decreased from C1 to C3 so that the three data classes can be easily distinguished. The first three isometric transformations, translation, rotation, translation, from the first stage of the method were applied to the three data classes (Figure 9(B)). The result is a mapping of the (x, y) segments from Figure 9(A) into a different plane (Length, Intercept) in Figure 9(B). In the process, the data classes were made parallel, and horizontal, i.e., zero slope. Another way of looking at the first stage of the analysis of the method is a sorting of the data classes by the intercepts, b, n, d, while normalizing across the slope (slope = 0).

[00169] The total length of the line segments in the region of interest is given by L={XProjection/cos[atan(Slope)]}, where the Xprojection is 10 in this particular case (Figure 9(A)), and represents the x range, while Slope is the slope of the fit lines, i.e., a, m, c. To illustrate the total Length (L), the three values for the three data classes were plotted with a square, circle, and triangle on the right hand side in Figure 9(B). The length (L) depends first on the slope (Figure 9(C)), but also on the X projections/X range (Figure 9(D)). For another illustration of the method, the points, (e.g., square, circle, triangle) were plotted for the three data classes shown in Figures 9(A) and (C). The distance along this curve between data classes C1 (square) and C3 (circle) is proportional with the dynamic range, D, which is defined in the second stage of the method (see Figure 1 (D) and Figures 4(A) - (B)), and which was used for ranking the five FBS lots.

[00170] In the process of untangling the data by the first stage of the method (see Figures 9(A) - (B)), the separation between data classes increased exponentially (see Figures 9(E) - (F)). For example, the segments with the smallest symbols are very close to each other in Figure 9(A), but have a larger separation in Figure 9(B). The first stage of the method amplifies the distance between the segments as shown by Figures 9(E) - (F), and this amplification increases exponentially as the intersection point of the lines is approached. With the amplification of the distance between the data classes, the noise is also amplified by the first stage of the analysis. In some embodiments, the method disclosed herein cannot operate beyond the discrimination limit, d-limit (see Figure 8(B)). Therefore, the exponential amplification stops at the d-limit which is equal to 9.8 for this FBS lot.

[00171 ] The separability between the classes of patients, as defined in the methods section as a Sep function, increases at least two orders of magnitude as a result of exponential amplification produced by the first stage of the method. In some embodiments, the percent increase of the Sep function, after applying the first stage of the method for the five FBS lots, is in the range 200 to 2500%. Furthermore, the overlap probability between classes of patients as measured by the t-test decreases several orders of magnitude after applying the first stage of the method.

Example 12: Classification of Clinical Trial Data

[00172] In some embodiments, the method disclosed herein may also be used to enhance the separation of clinical trial data. For example, Figure 1 1 shows the change in mini-mental state examination (MMSE) or Folstein test after three hours of administering Bryostatin or Placebo versus the change in Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) at 48 hours. Figure 1 1 (A) shows the dependence is linear for both Bryostatin and Placebo. In some embodiments, the isometric transformations disclosed herein may be applied to clinical trial data to change the two groups as shown in Figure 1 1 (B). Based on the average values of the Change in MMSE (Figure 1 1 (C)) plus standard errors of the means, there is some overlap and the separation is statistically significant under Mann-Whitney Litest (alpha 0.10, Figure 1 1 (E)) as well as under Wilcoxon test (alpha 0.10, data not shown) or t-test (alpha= 0.097, data not shown). After using the isometric transformations disclosed herein, the separation is definite (Figure 1 1 (D)) and is statistically significant under the Mann-Whitney U-test (any alpha i.e., ideal case, Figure 1 1 (F)).

[00173] Example 13: Classification of Genetics and Astronomy

[00174] In some embodiments, the method disclosed herein may also be used to enhance the separation of e.g., raw data for two cell cycle-regulated genes of yeast Saccharomyces cerevisaie and rotation curves of low surface brightness galaxies. Spellman P.T. et al., Comprehensive Identification of Cell Cycle-Regulated Genes of the Yeast Saccharomyces cervisiae by Microarray Hybridization, Mole. Biol. Of the Cell 9, 3273-3297 (1998); Kuzio de Naray, T., McGaugh, S.S. deBlok, W.J.G. & Bosma, A., High Resolution Optical Velocity Fields of 1 1 Low Surface Brightness Galaxies, ApJS 165, 461 -479 (2008); Kuzio de Naray, T., McGaugh, S.S. deBlok, W.J.G. & Bosma, A. Mass Models of Low Surface Brightness Galaxies with High Resolution Optical Velocity Fields, ApJ 676, 920-943 (2008). Figure 10(a) plots the rate data for the two cell cycle- regulated genes of yeast Saccharomyces cerevisaie; Figure 10(c) is the algorithm optimized data for two cell cycle-regulated genes of yeast Saccharomyces cerevisaie. Figure 10(b) plots the rotation curves of low surface brightness galaxies the velocity versus the radius for eight low surface brightness galaxies shows a linear dependence near the origin for eight data classes (C 1-8 ); Figure 10(d) is the algorithm optimized data that shows what the observed velocity of the galaxies would be if the radius of the galaxies would be 70 arsec.

[00175] Example 14: Age Dependence fo Alzheimer's disease Diagnostics

[00176] Figure 12 illustrates the validation of the method for resolving the age dependence for Alzheimer's disease diagnostic assay. In Figure 1 2(a), the raw data is presented showing the linear dependence of the assay output, Ln(A/N), for the two data classes, claims 1 -AD-squares and class 2-AC circles, the age of the patient. Figure 12(b) shows the algorithm normalized data for the same two classes of patients.

[00177] Discussion

[00178] In some embodiments, the method disclosed herein applies to increasing diagnostic separation for two or more linear classes of patients, (Ci , C 2 ,..., C n ), which change with respect to a second discreet variable, y, and therefore change the cutoff values. In some embodiments, the first stage of the method normalizes the data classes with respect to the slope, i.e., slope = 0, sorts data classes by the intercepts, establishes a fixed cutoff, and makes the data classes equidistant with respect to the fixed cutoff. In some embodiments, the second stage of the method ranks the output with respect to the second variable, y, e.g., FBS lot. The ranking is based on a measure of the dynamic range, D, and on the signal to noise level, <CV>/D, is called reference standard, D=D(Rank y ). As discussed above, this reference standard is linear even when the pairs of data classes are randomly generated.

[00179] In some embodiments, the methods disclosed herein may be used with a diagnostic assay requiring ranking of the separation, D, e.g., between AD and age- matched controls, when the second variable (e.g., FBS lot) changes this separation. As discussed above, the method disclosed herein was used with five lots of Fetal Bovine Serum (FBS) from three companies and with randomly generated data classes to build a reference standard, D=D(Rank y ). In some embodiments, the reference standard may be used for classification of any unknown FBS lot. The first stage of the method was also validated with a different assay measuring the network complexity and generalized for three data classes AD patients, Age-matched controls (AC), and Non-ADD patients. However, in some embodiments, the same first stage can be generalized to a larger number of data classes. The method was also tested with randomly generated samples and with three classes of patients (Figure 6 and Figures 7(A) - (B)).

[00180] In some embodiments, the first stage of the procedure applies to separating any classes of patients, (C 1 ; C 2 ,..., C n ) (see Figure 7), depending linearly on variables (e.g., hidden or unexpected). These data classes may be chosen from machine learning, neural networks, data mining, gene expressions, pattern or face recognition, cognitive psychology, or astronomy. To demonstrate the generality of the two stage procedure, two randomly generated data classes, as represented by circles, squares, and triangles, were considered from exponential functions of the type a * exp(-x/b), where a and b are parameters (see Figure 7(A) ; empty symbols represent the upper data sets and filled symbols represent the lower data sets).

[00181 ] In some embodiments, the method disclosed herein produces certain results as long as the intercepts of the data classes to be separated are statistically significant. In various embodiments, the method also produces certain results as long as the signal to noise ratio (SNR) is large enough not cross the separation boundary/cutoff line (see Figure 8(A)), i.e., on the left hand side of the discrimination limit, (d-limit). For a small signal to noise ratio, i.e., on the right hand side of the discrimination limit, (d-limit) (see Figure 8(B)), the method can be applied only for estimation purposes.

[00182] In some embodiments, the linear dependence on the first variable, x, needs to be established. From the practical and validation point of view, a linear dependence may be manipulated via isometric transformations, if compared with a nonlinear dependence. Therefore, this method determines in which representation the dependence on the first variable, x, is linear. If the initial dependence is nonlinear and can be assumed exponential, then through transformations (e.g., a natural logarithmic plot), the dependence can be linearized. This process is often called "linearization." For one of the AD diagnostic assays disclosed herein, the plot of the Ln(A/N) rather than A/N reached this linearization when represented versus Ln(Cell Density). The second AD diagnostic assay disclosed herein also has a linear dependence (see Figure 6). A linear dependence/input-output function is also desirable for the predictability of any system in general because a small perturbation in the independent variable/input of the system may produce a predictable and linearly correlated small perturbation in the dependent variable/output. [00183] Conclusions

[00184] In some embodiments, the two-stage procedure of the method disclosed herein may be applied to an AD diagnostic assay for which the output depends on two variables (e.g., cell density the experiment and FBS). The dependence of the assay output on one of the variables is linear. The second variable changes the parameters, i.e., slope and intercept, of the linear dependencies for each class of patients/data.

[00185] In some embodiments, the first stage of the method disclosed herein normalizes the linear dependence across the slope (slope = 0) making the classes of patients/data parallel, and equidistant with respect to a fixed Cutoff. Within the process of untangling the classes of patients/data, the distance between the classes of patients increases exponentially. The signal-to-noise ratio establishes the discrimination limit (d-limit) as the location where the two classes of patients are crossing over the Cutoff boundary, defined analytically herein as the equation of the angle bisector.

[00186] In some embodiments, the first stage of the method is also applied to an AD diagnostic assay that distinguishes between three classes of patients: Alzheimer's Disease, Age-matched controls, and Non-Alzheimer's Disease Demented patients.

[00187] In some embodiments, the second stage of the method ranks the normalized classes of patients by the distance and the normalized coefficient of variation. This standard curve is linear, D=D(Rank y ).

[00188] In some embodiments, to emphasize the generality of the linear standard curve for other AD diagnostic assays, D=D(Rank y ), pairs of random data classes were used. Table 1 : In-depth demographic, genetic/family history and clinical history of the Banked patients.

* Samples were collected under blind conditions.

Table 2: Information regarding the Fresh Clinic Samples

66