Field of the Invention

This invention pertains in general to the field of cancer diagnostics and cancer treatment. More particularly the invention relates to a multi-modal method for the profiling of a cancer tumour to the prediction of drug treatments to Malignant Melanoma by techniques such as molecular expression, digital pathology, Al-based imaging and classification on single cell level in malignant tumor Melanoma Tissue, a blood sample, or any other biofluid, aligned with both gene, protein, and/or metabolite expressions. The method and principle has a generic nature and can be applied to in principal to any type of solid tumor cancer originating from the skin and from other solid tissues, but not restricted to a particular cancer diseases, such as lung, colon, pancreas and prostate.

Background of the Invention

It is known that the ever-increasing demands in health care today position high forecasts and commands onto the research community to launch and come up with new solutions that can improve clinical outcome with better-quality and cost efficiency. In rejoinder to these contests, modern healthcare is looking for ways to treat patients that are both more effective as well as more cost redeemable. The introduction of regulatory directives are central to our research municipal in order to accomplish and meet the demands from e. g., cancer patients that are expecting drugs that are more safe, with lesser mortalities, and with a fast commencement of efficacy.

Melanoma malignancy is also known as malignant neoplasm or pathological malignant tumor expansion of skin-pigmented cells. The cancer disease mechanisms constitute an abnormal cell growth that will consequence in the invasion and most conceivable spread.

From the global Melanoma frequency, it is estimated, that the incidence of melanoma falls in the 19th place among cancers (GLOBOCAN database). However, in Europe, the ranking is on the 6th position. In recent patient number cases, of all diagnosed cancers in Europe (in 2020), 150 627 (3.4%) were melanomas. In addition, 26 360 of these patients, representing 17.5% of all diagnosed melanomas, lethally relapse following treatment. Metastasis progresses by cancer cell transitions to other parts of the organs and regions within the body, originating as a leak from the solid tumor origin. The most shared and signs of indications usually includes growing regional lymph nodes, loss of weight, general weakness and specific signs of organ dysfunctions such as neurological deficits.

Recently the cancer research area has been going through a progressive change of developments where the genetic variety of Melanoma classes of cancer, defined by the World Health Organization “WHO”, defines itself through the characterization of the disease pathology. Tumor heterogeneity is very commonly identified in Melanoma patients, with an extensive genetic alteration within a given tumor. These new findings on disease presentation opens a possibility to discover drugs that have not shown to have expected treatment advancements, or, within cases with a given particular genetic profile.

Therefore, there is a need for methods for characterization of the disease pathology and for linking the characterization to drug response and patient treatment.

Summary of the Invention

Accordingly, the present invention preferably seeks to mitigate, alleviate or eliminate one or more of the above-identified deficiencies in the art and disadvantages singly or in any combination and solves at least the above mentioned problems by providing a method for profiling a cancer tumour, wherein the method comprises the steps of; in a tissue sample from a cancer tumour, 1) Preparing at least one tumour slide from histological preparations; 2) Acquiring an image of each tumour slide, preferably correlating image pixels to slide coordinates, through digital pathology; 3) determining tumour heterogeneity by digital pathology analysis of each sample image from step 2, and classifying the tumour into different heterogenic sections; 4) acquiring a sample of each heterogenic section determined in step 3) from each tumour sample slide; 5) determining the genetic heterogeneity for each heterogenic tumour section as by molecular analysis on a protein or DNA level of each heterogenic section sample acquired in step 4).

Also is provided the method where step 3) further comprises; i. An image preprocessing step, wherein image inadequacies are corrected; ii. An image segmentation step where cells or groups of cells of interest are labelled; iii. A phenotypic analysis for single-clone selection, wherein phenotypic class and probability values are assigned to each labelled object; and iv. A Contour collection and transformation step, wherein all or top candidates are selected as heterogenic sections.

Further is provide a method for profiling the cancer burden in an individual, wherein the method comprises the steps of, 1) profiling at least one cancer tumour in the individual using the method for profiling a cancer tumour; and 2) optionally, profiling possible lymph node metastasis or circulating tumour cells (CTCs) [from blood] using at least step 5) and 6) of the method of any one of claims 4 to 10, and determining the cancer subtype of each profiled metastasis or CTC.

Also, is provided that the method for profiling the cancer burden further comprises the steps of the ability to generate a phylogenetic tree of the tumors’ evolution based on the data from each subtype and correlating the phylogenetic tree of the tumors’ evolution to the tumor progression; and Identify the driving cancer subtypes.

Further, is provided a method that may predict the treatability of a cancer burden in a patient by a medicament or a cocktail of medicaments, wherein the method comprises, 1) profiling cancer burden in an individual using the method for profiling the cancer burden; 2) for each identified driving cancer subtype, determining the cancer subtype potential drug response for known cancer drug treatments; and identifying if any known drug will likely have effect on all cancer subtypes, and/or identifying if any known drug that will likely have effect on driving subtypes, and/or identifying if any known drug that will likely have effect on the main driving cancer subtype, and subtypes with co-driving functions.

Also is provided a method that may select a medicament or a cocktail of medicaments for the treatment of a cancer patient, wherein the method comprises, 1) profiling cancer burden in an individual using the method for profiling the cancer burden; 2) for each identified driving cancer subtype, determining the cancer subtype potential drug response for known cancer drug treatments; 3) selecting a drug that targets all cancer subtypes; selecting a drug that targets the most lethal subtype; or selecting a drug that targets the most lethal subtype and selecting a second drug that targets the second most lethal subtype to prevent tumour proliferation while the most lethal subtype is being treated.

Also is provided a method for treatment of a cancer patient by a medicament or a cocktail of medicaments, wherein the method comprises, 1) Selecting a drug or group of drugs using the method for selecting a medicament or a cocktail of medicaments, 2) determining the individual patient maximum tolerance to treatment; 3) if a single drug is selected, calculating the dose of based on the patient tolerance, or 4) If at least two drugs are selected, calculate a high dose for the treatment of the most lethal subtype, and a low dose for the second most lethal subtype, wherein the high dose is 70-90 wt-% and the low dose 10-30 wt-% of the total dose based on the patient tolerance; 5) treating the patient using the medicament of cocktail of medicaments of step 1) and the dose regiment of step 3) or 4).

Brief Description of the Drawings

These and other aspects, features and advantages of which the invention is capable of and will be apparent and elucidated from the following description of embodiments of the present invention, reference being made to the accompanying drawings, in which

Fig- 1 is a schematic outline of the drug treatment concept of the invention;

Fig- 2 is a workflow for Molecular Profiling of Tumor Samples for the invention that incorporates the molecular pathology of expression, as well as cellular morphology on a single cell level annotation, aligned with the pathological manifestation of the tumor morphology and characteristics ;

Fig- 3 shows the identification of mutated variants of NRAS and WT NRAS by mass spectrometry. (A) Assigned MS/MS spectrum of the TMT-labeled peptide QVVIDGETCLLDILDTAGK corresponding to the mutation NRAS Q61K.

Fig. 4 shows the identification of mutated variants of NRAS by mass spectrometry. Assigned MS/MS spectrum of the TMT-labeled peptide QVVIDGETCLLDILDTAGR corresponding to the mutation NRAS Q61R.

Fig. 5 is a slide showing the signaling pathways within Melanoma cancer, where one or multiple proteins within the pathway are mutated, post-translationally modified, and/or present an up/down-regulation expression within the cancer stage, (intratumoral genetic heterogeneity in as sample from a Melanoma patient, presenting itself with heterogenous tumor clones within the tumor);

Fig. 6 shows Histological indices used to group of tumor samples, here histopathology staging and scoring of hematoxylin and eosin stained in combination with immunohistochemistry is used for precise groupings of patients;

Fig. 7 shows heterogenous tumor clones within the tumor of a Melanoma patient, in the invention used for the determination of intratumoral genetic heterogeneity; Fig- 8 shows a resulting chromatogram performed by liquid phase nanoseparation, interfaced to mass spectrometry, and how protein ID delivers functional attributes by mass spectrometry sequencing, that was not predicted by gene sequence ID alone;

Fig- 9 shows examples of resected lung tissue from Cancer patients, undergoing Drug treatment following the characterization method of the invention;

Fig. 10 shows an example of Computed tomography image generated from lung cancer tumor within the pulmonary tract;

Fig- 11 shows an example of a patient tumor being analysed by digital pathology, aided by Artificial Intelligence (Al) compartment index settings. Based upon the hallmarks of the tumor with its respective areas and compartments, the regions are denamed and annotated;

Fig. 12 shows an example of a primary tumor with 5 Melanoma tumor subclones were identified, with its respective histological features and morphology structures provided in the lower part; numbering relating to the clone number;

Fig. 13 shows (upper left) single cell-based segmentation from Subclone 2 region using deep learning-based image analysis method. In the upper-right is shown phenotypic classification of the subclonal regions at a single cell level, shades of gray decode phenotypic classes. In the bottom left is shown a confusion matrix of the used single-cell classification algorithm using 10-fold cross validation of five cancerous and two normal cell types. In the upper right is shown the phenotypic composition of the subclonal regions;

Fig. 14 shows the outcome of a compartment protein expression profile, where the specific PTM, including mutations are determined by LC-MS. The isolation of the generated Mass Spectrum (MS data) is shown, and further analysis of the protein made by protein sequencing and protein identifications (MS/MS data) by analysis that generates the amino acid sequence of the proteins;

Fig. 15 shows an example of a typical heterogeneous patient tumor, where the three marked regions are magnified in a digital slide scanned at 20X magnification (displayed at lower resolution). The regions of interest are exposed on the right side hand in a higher magnification on the right alongside tumor compartment scores provided by an expert pathologist;

Fig. 16 shows H&E generated from a patient tumor that has disease presentation that will alter the organ function of the lung, as presented herein where multi -factorial disease stages are driving the tumor progression and growth; Fig. 17 shows BRAF V600E mutation on DNA level with direct sequencing in surgical specimens from patients. The left graph shows the wild-type sequence. The right graph shows the BRAF V600E mutation with the T1799A transversion, as typical from a cancer patient;

Fig. 18 illustrates a molecular profiling by differential protein expression analysis, as shown in 2 resulting outcomes; control tissue (upper spectra), and Tumor tissue (lower spectra), with clearly visible expression differences in expression;

Fig. 19 shows a High resolution Computation Tomography, aligned with MRI and PET, and in combinations thereof, are imaging technology platforms of major importance for the Cancer surveillance and diagnosis of Cancer patients, with a tumor alteration as shown with targeted drug impact after 15 days;

Fig. 20 illustrated how selective inhibitors of EGFR-Tyrosine Kinases can alter and block the signaling by a small molecule drug, or by antibody bindings, by entering into the tumour cell and directly inhibits the EGFR-TK enzyme, regardless of the activation mechanism, thereby inhibiting the signaling events triggered by EGFR-TK activation;

Fig. 21 shows resulting expression profiles on a functional level by deep sequencing of proteins within the cancer tumor will ultimately generate an expression profile, whereby the pathway activity and mechanisms can be elucidated and outlined from the annotations made into pathway schemes of action, within lung cancer patients;

Fig. 22 shows how in order to handle the large data sets, a selection is made where the most prominent, and important information is gathered and selected. Experimental data are entered into a multi-dimensional evaluation system, that captures the main aspects of the data output and offers a grading of the respective and selected target candidate, based on expression/quantitation/PTM; and

Fig. 23 shows an illustration of the multi-drug cocktail treatment workflow with the follow up procedure that is acted upon after treatment response.

Description of embodiments

The following description focuses on an embodiment of the present invention of a multi-modal method for the characterization of a cancer by molecular expression in malignant tumor tissue, a blood sample, or any other biofluid, aligned with protein expressions. The method focuses in particular on prediction drug treatments to Malignant Melanoma, however, it will be appreciated that the invention is not limited to this application but may be applied to other solid cancer types.

In particular, the invention describes a multimodal methodology, capable of automatically predict the Drug treatment within Melanoma cancer samples, by mass spectrometry annotations of key regulatory molecules, preferably proteins, in combination with clinical data, including digital pathology, that provides a scoring for preferable drug, or multiple drugs provided as single therapeutics or in combination as a drug cocktail. Disease presentation characteristics based on a genotype-, and subsequent functional expression by proteins forms the basis of a particular phenotype that will drive the cancer tumor development, and this can be elucidated by the methodological steps of the HOPE-AI method. In addition, HOPE- Al, is a method based on the tumor specific diagnosis that also may be based on Expression. Subtypes may be determined based upon the HOPE method , and also subclones manifested based on the clinical scoring, molecular expression and Al based image analysis, and this may be used to determine an individual and optimized drug treatment therapy.

Malignant Melanoma

Usually, the start of the disease (Malignant Melanoma) is initiated from a pigment-containing cell, known as melanocyte. The most common factor in cutaneous melanoma development, relates to skin exposure to the environmental UV light. In these cases, the disease onset risk increases, when combined with low pigmentation of the skin. When large number of pigmented nevi has developed, usually other genetic and environmental factors are aligned within the disease mechanisms and evolvements. Usually, this is found in coordination with a compromised immune system, which can all underwrite and initiate melanomagenesis.

Pathological characterization is the main prognostic disease determinator for melanoma. The general schemes of the histopathological diagnosis and reporting has remained unaltered for several decades. It focuses primarily on the absolute tumor thickness, measured in millimeters (Breslow thickness). The Clark-level determination is a staging scheme that defines the depth of the existing skin anatomic structures involved by the melanoma growth. The Clark level of staging is structured according to the following level scoring;

Level I; confined to the epidermis (in situ) Level II; papillary dermis Level III; papillary and reticular junction

Level IV; — reticular dermis

Level V; — subcutaneous fat

The Breslow absolute depth is quantified by the AJCC8 classifications scheme (ref), which measures the vertical expansion the viable and identifiable melanoma cells from the top of the granular layer (undersurface of the stratum corneum) or from the top of the ulcerated tumor surface to the deepest viable melanoma cells.

Tis - melanoma cells confined to the epidermis mm; pTlb > 0.8 mm, but < 1 mm)

* T3 - 2-4 mm

* T4 - > 4mm

The ulceration is a significant upstaging factor, marked as ‘b’.

The Melanoma drivers, with their respective known mutations, has been identified; BRAF, NBAS, KIT, NF1, GNAQ, TP53, TERT, WNT/p-catenin, and the PTEN, as expressed genes, as well as proteins.

These markers cannot act as individual pointers for every melanoma case. In addition to the mutation status at the gene level, the actual amount of mutated protein has a critical influence on tumor biology and ultimately on patient survival. As Melanoma holds the highest genetic heterogeneity. This tumor structure will ultimately results in a high variability in the presenting disease phenotypes. In addition, the high level of heterogeneity represents a key element for the progression of the disease and has been linked to short survival.

Mechanisms involved in Melanoma that impacts the disease complexity is aligned with factor such as; metabolic shift from oxidative phosphorylation (OXPHOS), a preference for the glycolytic phenotype. However, there is an increasing number of studies linking mitochondrial pathways, particularly OXPHOS, to cancer development and progression, indicating an important Melanoma driver mechanism. In this way, the mitochondrial metabolism has been reported to impact cancer development in multiple ways, such as, increased generation of Reactive Oxygen Species (ROS).

Drug Treatment Concept In figure 1 is illustrated an insight and overview to the ever increasing number of melanoma patients and high risk of developing recurrent or metastatic melanoma, where the second primary tumor increases complexity of disease state, depending on site, and type of second primary cancers in melanoma patients.

This is aligned with the diagnosis and treatment opportunity for the patient. Type of therapy (treatment) that includes second-line treatment with a possible previous chemotherapy. Primary sites can be any type, e.g.; acral, mucosal, cutaneous, uveal, and unknowns, having disease-free survival (DFS), progression-free survival (PFS), as well as varying median overall survival (OS).

These patients also have an individual disease presentation within the tumor tissue. Digital images provides detailed characteristics that relate to the morphology as such, that is identified by the cellular context, in addition to the molecular morphology, whereby these cells are built from. In this context, the cell morphology will describe features such as; shape, structure, form, and size of these cells that are important building blocks of the tumor.

The morphology of the tumor is also strongly influenced by the microenvironment of the cell and cell compartments. The tumor tissue and tissue microenvironments will also be impacted by its response to biophysical and topographical cues is governed by mechano-sensing-, mechano-transduction-, and mechano-responses. The third pillar of the drug treatment fundament of the concept relates to the key regulating proteins in Melanoma and their specific active and functional proteoform they have within the disease setting. These active forms of the disease drivers and passengers are in most cases related to a cellular transformation that will change its form by; post-translational-modifications, such as, but not restricted to phosphorylation, acetylation, glycosylation, and methylation. The actual quantity of any given Melanoma key regulating target protein is also of mandatory importance, impacting disease progression and developments.

Invention workflow

In the invention, a workflow has been developed whereby data can be generated to characterize a cancer tumor and a cancer in a patient and provide data for successful drug treatment process. An example workflow can be seen in figure 2. The invention is exemplified using melanoma cancer, since Malignant Melanoma is one of the most aggressive types of cancer, of all cancer diseases in addition to the most aggressive skin cancer. In one embodiment is provided a method for profiling a cancer tumour, wherein the method comprises the steps of; in a tissue sample from a cancer tumour,

1) Preparing at least one tumour slide from histological preparations;

2) Acquiring an image of each tumour slide, preferably correlating image pixels to slide coordinates, through digital pathology;

3) determining tumour heterogeneity by digital pathology analysis of each sample image from step 2, and classifying the tumour into different heterogenic sections;

4) acquiring a sample of each heterogenic section determined in step 3 from each tumour sample slide;

5) determining the genetic heterogeneity for each heterogenic tumour section by molecular analysis on a protein or DNA level of each heterogenic section sample acquired in step 4).

The method may further comprise the step of;

6) Through bioinformatics analysis of the molecular analysis result in step 5), determining the number of different intratumor subtypes in the tumor sample of step 1.

In the method, the tumour may be a primary tumour, a metastasized tumour, or a primary tumour being melanoma.

Below the different method steps will be described in more detail.

Mass Spectrometry Technology and Methodology

Mass spectrometry (MS) is nowadays the most valuable analytical technology platform for clinical protein science, because it measures an intrinsic property of any given molecule, its mass, with very high sensitivity. MS can therefore be used to measure a wide range of molecule types and a wide range of sample types/biological materials.

In this ionization process, the precursor ion is activated by acceleration into a mass-selective linear ion trap under conditions whereby some of the fragment ions formed are unstable within the trap. After a time delay the stability parameters of the ion trap are changed to allow capture of fragments that that were previously unstable. The result is a product ion spectrum that originates from precursor ions with a modified internal energy distribution. It is possible to follow the evolution of the precursor internal energy distribution for many milliseconds after admittance of the precursor ions into the linear ion trap. Time-delayed fragmentation product ion spectra typically display reduced sequential fragmentation products leading to spectra that are more easily interpreted. Several important experimental parameters important to time-delayed fragmentation have been identified and are discussed. The technique has applications for both small precursor ions and multiply charged molecules. Tandem mass spectrometry (MS/MS) is at the heart of most of modem mass spectrometric investigations of complex mixtures. The fragmentation involves activation of a precursor ion via collisions with a target gas and may produce charged and neutral fragments. The nature of the fragment ions, as well as their intensities, is often indicative of the structure of the precursor ion and thus can yield useful information for the identification of unknown analytes, as well as providing a useful screening technique for different classes of analytes. Activation via multiple collisions both prolongs the activation time and enables higher energies to be deposited into precursor ions. Higher collision gas pressures also imply higher collision relaxation rates.

In figures 3 and 4 is shown the Identification of mutated variants of NRAS and WT NRAS by and Identification of mutated variants of NRAS, respectively, by mass spectrometry.

Digital Pathology

The tissue sections can also undergo histological staining after the investigation is completed. This is performed in order to target areas of interest, image scanning and capture. Within this context, digital pathology in diagnostics is an emerging and upcoming field that became a sub-field of pathology focusing on data management based on information generated from digitized specimen slides.

All through the use of computer-based technology, digital pathology exploits virtual microscopy, by having the glass slides with the patient tumor converted into digital slides that can be analyzed, by virtual microscopy. The workflow that provides the data input for a given patient tumor characteristic, constitute the following parts; The capability of transmitting the digital slides over distances quickly, which enables tele-pathology scenarios.

The capability of accessing past specimen from the same patients and/or similar cases for comparison and review

The capability of comparing diverse areas of multiple slides simultaneously with a virtual microscope.

The capability of annotating the areas directly in the slide and share this for research annotations. By immunohistochemistry, we are able to provide additional tumor details on key regulating protein targets that includes data on the immune phenotyping in tumors, biomarker expression profiling across the section.

In figure 6 is shown Histological indices used to group of tumor samples, here histopathology staging and scoring of hematoxylin and eosin stained in combination with immunohistochemistry is used for precise groupings of patients.

Multi-sectional preparations may be prepared, thus generating a cross sectional slide library across the tumour, enabling analysis of the entire tumor-heterogeneity.

Each patient undergoes a clinical clustering according to the standardized clinical characteristics as sex, age, location, and Fitzpatrick skin type. The survival data are stratified by DFS, PFS and OS parameters. The histopathological dataset includes the WHO 4th types, and the absolute thickness of the melanoma. The whole tissue section is the basis for deeper morphometry of the melanoma tissue for subclones, heterogeneity, and stromal phenotyping. The single cell identification of the certain cells or sublones is encouraged by immunohistochemistry, which also serves for tissue validation of the tissue protein biomarkers, together with the survival data for biomarker relevance assessments. These data can be summarized as clinical progression and histopathological adverse scales.

The tumor heterogeneity can also be determined by digital pathology and different heterogenic section and generation of samples through laser dissection. Fig. 7 shows heterogeneous tumor clones within the tumor of a Melanoma patient, in the invention used for the determination of intratumoral genetic heterogeneity.

Melanoma Tumors are most commonly heterogeneous in nature and constitute the disease progressive driving cell phenotypes, which causes the patient to experience a tumor burden increase over time within the disease evolvement. An example of a typical heterogeneous patient tumor is outlined (shown in Fig. 15), where the three marked regions, are magnified in a digital slide scanned at 20X magnification (displayed at lower resolution). The regions of interest are exposed on the right side hand in a higher magnification on the right alongside tumor compartment scores provided by an expert pathologist. These heterogeneous images builds onto our library, and forms the basis for an Al-driven algorithm build that is used within the method to align tumor types and cancer cell phenotypes within, that forms the nature of the tumor that needs to be treated.

The nature of the tumor will also be determined by the fact that the molecular changes and alterations due to the tumor impact will have an effect on the tissue function, and corresponding organ of that tissue. The main driver as in the case shown above (See Fig. 16), is from a Melanoma tumor within the central lobe of a patient lung, where additional functional alterations arise, such as a tumor progression, aligned with a heavy interstitial inflammation, a hyperinflation and a vascular remodeling within the lung parenchyma.

Al-based algorithm developments with Singe-Cell annotations

From a slide to an isolated cell. Below is shown an example of the workflow:

The major steps of the workflow

Input: Sample prepared for imaging

Steps:

- Data acquisition

Image preprocessing

Image segmentation

- Phenotypic analysis for single-clone selection

Contour collection and transform

Single cell/region isolation

Output: Isolated single clones

Thus, in the method of the invention, the determining tumour heterogeneity by digital pathology analysis step may comprise; (i) An image preprocessing step, wherein image inadequacies are corrected; (ii) An image segmentation step where cells or groups of cells of interest are labelled; (iii) A phenotypic analysis for single-clone selection, wherein phenotypic class and probability values are assigned to each labelled object; and (iv) A Contour collection and transformation step, wherein all or top candidates are selected as heterogenic sections.

Data acquisition

The pipeline receives a tissue sample that is mounted on a holder and made ready for single cell isolation (eg. PEN/PET membrane slide, frame slide, membrane slide well plate) as discussed in (Mund et al. 2022; Brasko et al. 2018). The tissue width may be between 2-20 Um. Imaging may be done by fluorescent or brightfield high- throughput or tissue scanning microscopes at a typical 5-100x magnification. The holder contains 3 or more marker points that help in the navigation while correlative microscopy is performed. Image preprocessing

Image inequalities, illumination problems and noise may be removed from the images before further processing. For this step CIDRE or similar methods are used (Smith et al. 2015). In the case when the used imaging device cannot record whole slide stitched images, the stitching is performed in this step.

Image segmentation

During this step images are partitioned to background and potential objects of interest (that are cells or cellular groups with similar morphological properties). This partitioning is done by using deep learning based segmentation algorithms that are trained on manually annotated cellular outlines. Typically, the amount of training data ranges from a few thousand to millions of cells. The method used for the segmentation is nucleAIzer (Hollandi et al. 2020) a parameter-free segmentation method, but other methods may be used as well (eg. StarDist, CellPose, OmniPose, UNET). The output of this step is uniquely labeled objects that are cells or groups of cells on the images.

Phenotypic analysis

This step assigns phenotypic classes to the identified objects. The classes may be predefined by expert (eg. cancer, lumen, healthy regions, etc.) or automatically defined by clustering algorithms. For the classification, information of the cellular microenvironment may be included by cellular neighborhood or transforms like fisheye transformation to incorporate the surroundings into the machine-based decision (Toth et al. 2018, Toth et al. 2022) . The classification is done either by using feature extraction and statistical learning approaches (Piccinini et al. 2017) or by an end-to-end deep learning image classification network. The output of this step is a set of class and probability values for each cellular object. The ability to determine unmature-, from matured cells, as an important disease phenotype contributing diagnosis within the HOPE method may also be implicated as a drug treatment determinator, as it relates to the molecular expression profile of these cell types, characterized by the imaging Al.

This is clearly shown in figure 13, which shows single cell-based segmentation from a Subclone 2 region using deep learning-based image analysis method (upper left). Also is shown is phenotypic classification of the subclonal regions at a single cell level, shades of gray decode phenotypic classes (upper-right). In the bottom left is shown a confusion matrix of the used single-cell classification algorithm using 10-fold cross validation of five cancerous and two normal cell types. The phenotypic composition of the subclonal regions is shown in the upper right.

Contour collection and transform

Once the class assignment is performed, the operator can choose a clone for isolation. During this the operator has three ways for selection: (1) select all, (2) select top candidates, that will select only highest probability entities or (3) manual selection. In practice (2) is preferred for fully automated workflows and the combination of (2 and 3) if manual intervention is feasible. After the contours are selected for isolation, the sample is put to the isolation device and the coordinates of the selected contours are transformed based on the markers on the holder.

Single cell/region isolation

The regions are then automatically isolated into single-cell holder caps or multi-well plates. The operator may perform two additional steps, (1) manual or automated focus detection at each single object and (2) fine alignment of the contours, that may be necessary for micron precision isolation or in the case when the sample goes through nonlinear distortions such as drying. Samples are then collected into caps based on their phenotypic classes.

Drugability

Detailed Proteoform identity and status of key regulating proteins, in combination with digital pathology characterization, and clinical data- properties, related to available anticancer drugs that can potentially benefit melanoma patients, opens up a novel avenue of patient treatments. One of the most active pathways, related to key functions in Melanoma, Is the in the Mitogen-activated protein kinase, kinase- “MAPK”- pathway. In addition, Akt, also known as protein kinase B (PKB), of the serine-threonine kinases have a precarious role by largely impacting the signaling network. In these cellular events, Akt activation serves as a master switch for cellular signaling pathways. The Akt signaling pathway or PI3K-Akt signaling pathway, functions by signal transduction that endorses survival and growth in response to extracellular signals. Intracellular cross-talks and signaling occurs, whereby cellular functions drive the responses through downstream targets and interacting partners. In this way, the MAPKAP pathway, is influenced in oncological Melanoma mechanisms with the Akt, having important roles in cell survival, growth, proliferation, angiogenesis, vasorelaxation, and cell metabolism. In addition, activated Akt mediates downstream responses, including cell survival, growth, and most of all cell migration and proliferation. The phosphatase and tensin homolog, PTEN is a complementary tumour suppressor to AKT, which is often mutated or lost in cancer cells. Akt phosphorylates as many as 100 different substrates, leading to a wide range of effects on cells.

The Proto-oncogene c-KIT is the protein that constitutes the receptor tyrosine kinase protein, also known as tyrosine-protein kinase KIT, (CD117), but also runs under the name of mast/stem cell growth factor receptor (SCFR).

The varying classes of available Melanoma medicines builds on a number of varying pharmacological principles. Targeted treatment drugs are interacting with the Protein target that is a key regulating protein that is associated with the tumor development and/or the metastasis development. Examples of drugs related to the targeted principles are outlined in Table 1-3, where the drug inhibitor name and classes are provided, along with the drug target and the status. These tables constitute these drugs, both within the respective clinical development phases, where access is provided for patients, to participate, as well as the drugs as products, readily available, in most cases worldwide.

Table 1. Inhibitor drugs, and classes targeting the AKT signalling

Table 2. Additional kinase inhibitors

Checkpoint inhibitors, or immunotherapy drugs called, acts by the blocking of the checkpoint proteins, from affinity interacting with their associated proteins. By this principle it prevents the-off signal from being sent, and thereby allowing the T-cells to kill the cancer cells. These are outlined in Table 3.

Table 3. Biologies, targeting immune checkpoints, angiogenesis and other cytokine pathways Chemotherapy is another type of principle whereby cancer is treated. The drug principle is denamed; chemotherapeutic agent that actually stops cancer cells from dividing and/or growing. There are several categories of chemotherapeutic agents. Due to the mode of drug action, these compounds will target any cell undergoing growth and cell dividing.

Currently, there exist approximately 100 different chemotherapy drug agents, where some of these are presented in Table 4, which are most commonly used in Melanoma.

Usually, patients will get more than one type of chemotherapy at a time in order to help attack the cancer in altered ways. As Melanoma considered, DTIC (dacarbazine) monotherapy is used at usually at 2 ^nd or 3 ^rd lined, mainly palliative regimens.

Table 4. Chemotherapeutic agents

Several insulin-sensitizing drugs are able to trigger the AMP-activated protein kinase (AMPK), by inhibiting the mitochondrial functions. These drugs, such as metformin and others, will typically inhibit the ATP synthesis in mitochondria, in order to raise the AMP/ATP ratio in the process within the AMPK activation. Some of the Mitochondrial pathways inhibitors with details are presented in Table5.

Table 5. Mitochondrial pathways inhibitors Approximately, 80% of the melanoma tumors carries genetic alterations, related to at least one of the key players in the MAPK pathway, as shown in Fig. 5. Herein, the Mitogen-activated protein kinase kinase (often denamed as MAP2K, MEK, MAPKK), are all kinase enzymes that are readily involved in the cancer signaling, promoting the tumor growth, by phosphorylating the mitogen- activated protein kinase.

Molecular histopathology diagnosis

The patient tumor is analyzed by digital pathology, aided by Artificial Intelligence (Al) compartment index settings. Based upon the hallmarks of the tumor with its respective areas and compartments, the regions are denamed and annotated, as exemplified by a patient tumor in Fig. 11.

Heterogeneous regions are being localized where very often the cellular shape and size of tumor regions are mapped and annotated. Corresponding H&E histology images are being used to verify and match for instance the specific clone within a tumor. This is performed on a single cell level, and is then used within the Al-learning part, developing an algorithm that will aid in predicting the correct phenotype of a given clone. All of these structures and activated cell types that incorporates, stem cell, and/or stem cell like cells are built into an overall library, that is being populated over time. The size of library is growing over time, as the number of tumor tissues are being evaluated and added into the library, whereby the Al, and algoritms are therefore enriched and improved over time.

The cellular expressions and annotations of proteins and genes are then becoming a part of the sample processing strategy for deep analysis of the melanoma proteome by mass spectrometry.

In figure 12 is shown a primary tumor with five Melanoma tumor clonal identified, with its respective histological features and morphology structures provided in the lower part; numbering relating to the clone number.

Within this methodology, there is a molecular profiling part that builds onto the tissue characterization, as described above, where multiple types of solid samples from melanoma patients are being analyzed and evaluated, such as typically, but not restricted to; primary tumors, lymph nodes and distant organ metastases. Samples from each region of interest, where typically clones of particular interest, are next isolated by dissection and the undergoing sectioning. Multiple sections are typically used for Mass Spectrometry analysis and one section is prepared for histology to determine the tumor cell content. The protein expression profiles typically generated are shown in Fig 14. Specific driver and passenger proteins and genes are identified and identified as targetable treatment options with product medicines on the market.

Molecular analysis of clones

Example of molecular analysis of a dissected region/clone described under Digital pathology above, where the presence of tumor driving mutations is determined.

The mutation being determined on a DNA level, for instance in BRAF, where the mutation is changed from V->E in the 600 amino acid position will have a mRNA, as shown in Fig. 17, and corresponding protein synthesis, and expression alteration that is altering over time, which have the functional impact in the disease activity of the patient.

The BRAF V600E mutation determined by DNA sequencing of patient surgical specimens is the first step in diagnosing Melanoma. The DNA is isolated from the tumor wherafter sequencing is performed to determine the BRAF mutational status. In Figure 17, the left parts show the wild-type sequence. On the right part, the graph is showing the BRAF V600E mutation with the T1799A transversion. BRAF mutations are T to A transversions at bp position 1799.

The undergoes transcription, whereby the DNA of a gene serves as a template for complementary base-pairing, and mRNA is formed.

The mRNA molecule might in the next following steps, be translated to synthesize the protein molecule, that was encoded by the original gene. The actions by the Proteome (all expressed proteins at a given time point), as well as functional Melanoma actions by single proteins as key regulators in disease is of mandatory importance to the cancer progression, and the ability to treat the tumor by targeted drugs.

This results in a tumor characteristic in a way that some cancer cells within the tumor will be mutated to varying degrees, where others will be wilt type.

Figure 18 shows an example of two resulting mass spectra outcomes, that compares control tissue (upper spectra), and Tumor tissue (lower spectra). Herein, clear differences can be identified, even on a visual inspection, that gives an understanding of the differences in protein expression. Next step relates to the identification and verification as well as the selection of drugs that can be assigned to the treatment of the patient, based on these data. The analysis will provide upregulation/downregulation of pathways for the most prolific cancer subtype, that we can built on in a complete bioinformatics network output, delivering Enriched Biological Pathways in patients primary-, and metastasis tumors, respectively. In the respective protein expression profiles, before and after (illustrated as vertical arrows and horizontal arrows), it is possible to align the proteins expressed, and their mutations and altered post-translational-modifications “PTMs”, (vertical profile), and the changes that can be identified, as a treatment effect of the provided drug treatments (See Figure 22).

Examples of regulations that are observed are immune-related pathways with significance to up-, or down-regulations, which can be matched to histological observations. Additional examples of disease presentations within the tumors that can be evaluated by this method relates to lymphocyte infiltration, a key output that can be used as a functional decision making, whether its high/low; in relation to a possible treatment decision, or, as a resulting outcome of a drug treatment effect.

Thus, the molecular analysis step of the method for profiling a cancer tumour further comprises that for each heterogenic tumour section, the presence of tumour driving mutations, not restricted to but exemplified, such as BRAF, NRAS, KIT, NF1, GNAQ, TP53, TERT, CDK4, TGF-B, EGFR, and the PTEN, is determined by performing targeted sequencing validation.

The method for profiling a cancer tumour may further comprise the step 6) Through bioinformatics analysis of the molecular analysis result in step 5), determining the number of different intratumor subtypes in the tumor sample of step 1.

Step 6) may further comprise determining the presence of: (i) Aggressive clones, (ii) Metastatic clones, (iii) Potential clone vulnerabilities, (iv) Mutation for keyregularity target proteins in cancer diseases, (v) Mutation for key-regularity target proteins in cancer diseases where specifically drug compounds are available, (vi) Potential drug response, or (vii) Correlation immunohistochemistry, through bioinformatics analysis of each intratumour subtype.

It may further comprise generating a phylogenetic tree of the tumors’ evolution based on the data from each subtype, and/or identifying the driving cancer subtypes.

Imaging

An example of the tumor spread throughout the body illustrated by a DG

PET/CT full body exposure of a Cancer patient with Malignant Melanoma is shown in Figure 19, using of a patient with malignant melanoma who received immunotherapy, and subsequently responded to the drug treatment.

In this respect, positron emission tomography (PET) only uses small levels of radioactive radiotracers or as sometimes denamed; radiopharmaceuticals, that is used together with a special camera associated by a computer that is evaluating the organ and tissue functions. The identified changes at a cellular level, that PET identifies, also as an early onset of disease afore any other imaging technology.

Clearly, PET/CT is a potent central imaging modality for cancer imaging. Herein, the tumor diagnosis provides indication of the spread and is key in

Assisting the diagnosis and tumor status.

In essence, the staging of patients with a newly diagnosed malignancy, as well as the restaging following therapy and surveillance are disease progression steps where these imaging platforms are key.

With a mutation of Valine to Glutamic acid within a metastatic melanoma.

After four additional doses of ipilimumab, the patient had a complete response.

This exemplifies situations, where patients with no previous metastatic melanoma discernible disease at baseline alters the tumor burden, and after multipledoses of targeted drug treatments (small molecules, or monoclonal antibodies, see Table with drugs), as shown in the figure. Figure 19 also shows surface pictures of tumor burdened moles, and the protein expression pathways that drives the tumor progressions. These key regulating proteins, that also show an alteration to include mutations, as protein modifications, that will drive the signaling and thereby impact the tumor growth.

After multiple additional doses of targeted treatments, the patient had a complete response, as illustrated in the PET/CT Imaging figures.

These drug effects on tumor heterogeneity is spotted by the method, and in addition, passenger regulators are of importance that subsequently is supporting the key drivers such as BRAF, NRAS CDK4, and KIT.

As melanoma encompasses high somatic mutational burdens among solid malignancies, our method can identify the clinical significance of the vast array of protein/gene, both drivers as well as mutations that are being prevalent in melanoma as passengers, appearing under transforming conditions.

Integration with chromosomal copy number data contextualized the landscape of driver mutations, providing oncogenic insights in BRAF- and NRAS-driven melanoma as well as those without known NRAS/BRAF mutations, are mostly common. The method would also direct an activation by KIT aberration, predicting a response to tyrosine kinase inhibitors. Data on; imatinib, nilotinib, or dasatinib; and some NRAS-mutant tumors may exhibit sensitivity to MEK inhibition are of significance.

Patient Tumor Tissue Characterization and Scoring

The clinical categorization and scoring is what the tumor sample will build on based on the individual personalized characteristics, that includes; Clinical clustering, histopathological clustering, immunohistological clustering and the corresponding meta-data of the patient. Reference library establishment is made from these individual patient tumors (patient #1, patient #2, patient #3, patient #4, patient #n), that forms the basis for the Al-algorithms to be based on, and from where the algorithms and models will be generated.

Lung cancer

It is common that a cancer results in tumors in other tissue, such as lung tissue. Selective inhibitors of EGFR-Tyrosine Kinases can alter and block the signaling as depicted in Fig. 20 by a small molecule drug, or by antibody bindings, by entering into the tumour cell and directly inhibits the EGFR-TK enzyme, regardless of the activation mechanism, thereby inhibiting the signaling events triggered by EGFR- TK activation.

By inhibiting EGFR-TK, the drug impact and action will inhibit the key mechanisms of tumor cell growth and progression that are driven by the EGFR-TK. These events will result in decreased tumor growth and/or tumor shrinkage.

Resulting expression profiles on a functional level by deep sequencing of proteins within the cancer tumor will ultimately generate an expression profile (as shown in Fig 21), whereby the pathway activity and mechanisms can be elucidated and outlined from the annotations made into pathway schemes of action. In terms of protein expression, the exact contributions of specific cells within specialized tissue compartments are often difficult to resolve in tissue composed of heterogeneous cell types varying in activation and differentiation states. Within any cellular compartment, the relative concentrations of individual proteins vary, and the gradients of protein concentrations occurring in extra-cellular compartments will have an important impact on their potential functional activities. Clone Heterogeneity Characterization

By aligning the high resolution image to the Al-algorithm system, diversity of the cancer cells within a given clone can be made, where clone areas can be defined as a novel entity definition of the patient tumor and as a hallmark of the characteristics of the patient tumor on a Personalized level. The clone hallmarks in terms of Cancer cell type, as well as the presence of the Immune cells and Stromal part, defined by a ratio of; C:I:S (as exemplified by the pie charts presented in Figure 12), which forms the basis of the HOPE- Al method.

The image capture of the Al-algorithm output with an overlay of the H&E original image, gives landmark outputs in relation to the healthy cells within the patient tumor. This is also illustrated within the Figure, were the clones are depicted one by one, in relation to the diseased 5 different cancer clones.

The HOPE-AI method also is able to separate in-between the different clones, that provides evidence on differences in cell shape and size. The Nuclear phenotype is one important feature read, related to the size and shape of the cancer cells.

In figure 13 is shown single cell-based segmentation from Subclone 2 region using deep learning-based image analysis method (upper left). Also, phenotypic classification of the subclonal regions at a single cell level, shades of gray decode phenotypic classes (upper-right). Confusion matrix of the used single-cell classification algorithm using 10-fold cross validation of five cancerous and two normal cell types (bottom left). The phenotypic composition of the subclonal regions is shown in the upper right.

Predictive Treatment Concept Strategy

The outcome from a molecular deep sequencing analysis is a big-data result, that is a status report on the disease presentation. In order to handle the large data sets, a selection is made where the most prominent, and important information is gathered and selected. The functional aspect here is the proteome, i.e the protein expression at a given time point within the patient, a given tumor region (in primary of metastasis tissue), a special clone of the tumor, that has a given characteristic. These experimental data are next entered into a multi-dimensional evaluation system (see Figure 22), that captures the main aspects of the data output and offers a grading of the respective and selected target candidate, based on expression/quantitation/PTM. The cell type origin in terms of phenotype/context/reference, all of which is then determined as disease knowledge.

The DNA-RNA-Protein-Metabolite expressions of technologies empower and provide us with a variety of approaches for studying the molecular and biological bases of disease. We are able to study the dynamics of molecular expression within a variety of biological compartments ranging from sub-cellular sites to the entire tumor tissue. By understanding key pathways functioning in the maintenance of steady-state function, activation, proliferation, apoptosis, repair, growth, and regulated gene expression, we are discovering the fundamental regulatory mechanisms of disease, and the possible sites for medical intervention strategies. The assignment of protein identities which are linked to key biological mechanisms associated with disease processes and disease progression are an important area of activity. The unmet challenges of molecular expression activities today revolve around understanding the roles which individual Genes and Proteins or entire pathways play in the initiation and development of complex Cancer diseases.

There are basic points which need to be attained: (1) A betterment and outline of the general understanding of the pathophysiological basis of disease, being measured. (2) A linkage between clinical investigations which provides quality clinical samples for study and standardized clinical measurements for correlation to biological activity. (3) Dedicated searchable Databases of annotation for gene and proteins, both which are cross referenced to biological processes occurring in human disease and linked to data acquired using technology platforms. (4) A linkage between diseases which allows large datasets to be compared to common reference proteomes, be it cell or organ. This would include the development of statistical tools and approaches for comparing large datasets in unbiased testing. The development of technology which will allow us to break the “abundance barrier”

The illustration in Fig. 23 gives an overview of the method steps and process work flow. The patient will be given a diagnosis at the hospital, and supported by a line of treatment. At a given time point, the lack of response to these treatments, that usually positions the patient on a worsening of its health status. This will then result in that an in-depth Molecular Profiling and Mutation/PTM/Pathway screening on gene and protein level will be performed, where after computation and analysis dedicated target molecules will be determined. These proteins and genes will be characterized based upon their respective PTM, expression level, as well as pathway functional actions. A corresponding multi-drug cocktail will be by low dose administration, as a personal and dedicated follow up treatment, and as a complement to the guideline tretaments in Cancer. After a given time, the tumor burden status will be determined, most probably but not unconditionally by image analysis, where the add on effect of multiple low dose treatment is evaluated. Next, there is an opportunity to adjust the dose and frequency, whereby the treatment is given.

Thus, in the invention the cancer burden in an individual may be profiled by profiling at least one cancer tumour in the individual. Optionally, possible lymph node metastasis or circulating tumour cells (CTCs) (from blood) may also be profiled, whereby the cancer subtype of each profiled metastasis or CTC may be determined.

Further, a phylogenetic tree of the tumors’ evolution based on the data from each subtype may be generated. The phylogenetic tree of the tumors’ evolution may be correlated to the tumor progression. Importantly, driving cancer subtypes may be identified.

This way, the central determinator, that is the main driving cancer subtype may be identified, and subtypes with co-driving functions.

Pharmacology and multi-drug combinational treatments

In this respect, the key-regulated DNA-mRNA- and corresponding protein expression being annotated within the pathological staging of the patient will be mapped onto the pharmacology prediction database system developed, where the key regulated proteins will be mapped against drug compounds, may that be;

* small molecule drugs

*Monoclonal antibody drugs

*Antibody fragments, Fabs and others

*Cell therapies

*mRNA vaccinations

The resulting pairing of the Target protein(s)-drug medicine(s) will next be evaluated against the health status of the cancer patient, where risk factors of specific drugs, and combinations thereof will be taken onto consideration.

As possible “cocktail” of multi-drug treatment will be the option as a drug administration to reach “optimal and individualized” efficacy, the dose to be considered and determined, will be calculated as an outcome of the HOPE-AI method.

Combining these methods, the treatability of a cancer burden in a patient by a medicament or a cocktail of medicaments may be predicted. This may be done profiling cancer burden in the individual. For each identified driving cancer subtype, one may determine: the cancer subtype potential drug response for known cancer drug treatments; and identifying if any known drug will likely have effect on all cancer subtypes, and/or identifying if any known drug that will likely have effect on driving subtypes, and/or identifying if any known drug that will likely have effect on the main driving cancer subtype, and subtypes with co-driving functions.

This way, a medicament or a cocktail of medicaments may be selected for the treatment of said cancer patient. For each identified driving cancer subtype, the cancer subtype potential drug response for known cancer drug treatments may be determined, and a drug may be sleceted that: selecting a drug that targets all cancer subtypes; selecting a drug that targets the most lethal subtype; or selecting a drug that targets the most lethal subtype and selecting a second drug that targets the second most lethal subtype to prevent tumour proliferation while the most lethal subtype is being treated.

The selected medicament or a cocktail of medicaments may be used to treat the patient. However, a cancer patient may already be in a weakend state, why it is beneficial to determine the individual patient maximum tolerance to treatment, and

If a single drug is selected, calculating the dose of based on the patient tolerance, or

If at least two drugs are selected, calculate a high dose for the treatment of the most lethal subtype, and a low dose for the second most lethal subtype, wherein the high dose is 70-90 wt-% and the low dose 10-30 wt-% of the total dose based on the patient tolerance,

Whereby the patient may be treated using the medicament of cocktail of medicaments in the calculated dose regimens.

Preferably, the patient cancer burden is monitored for changes in the cancer burden, for instance through treatment effect or medical environment, such that selection of medicaments or dose may be adjusted. Although the present invention has been described above with reference to (a) specific embodiment s), it is not intended to be limited to the specific form set forth herein. Rather, the invention is limited only by the accompanying claims and, other embodiments than the specific above are equally possible within the scope of these appended claims, e.g. different than those described above.

In the claims, the term "comprises/comprising" does not exclude the presence of other elements or steps. Furthermore, although individually listed, a plurality of means, elements or method steps may be implemented by e.g. a single unit or processor. Additionally, although individual features may be included in different claims, these may possibly advantageously be combined, and the inclusion in different claims does not imply that a combination of features is not feasible and/or advantageous. In addition, singular references do not exclude a plurality. The terms "a", "an", “first”, “second” etc do not preclude a plurality. Reference signs in the claims are provided merely as a clarifying example and shall not be construed as limiting the scope of the claims in any way.

References

Brasko, Csilla, Kevin Smith, Csaba Molnar, Nora Farago, Lili Hegedus, Arpad Balind, Tamas Balassa, et al. 2018. “Intelligent Image-Based in Situ Single-Cell Isolation.” Nature Communications 9 (1): 226.

Hollandi, Reka, Abel Szkalisity, Timea Toth, Ervin Tasnadi, Csaba Molnar, Botond Mathe, Istvan Grexa, et al. 2020. “nucleAIzer: A Parameter-Free Deep Learning Framework for Nucleus Segmentation Using Image Style Transfer.” Cell Systems 10 (5): 453-58. e6.

Mund, Andreas, Fabian Coscia, Andras Kriston, Reka Hollandi, Ferenc Kovacs, Andreas-David Brunner, Ede Migh, et al. 2022. “Deep Visual Proteomics Defines Single-Cell Identity and Heterogeneity.” Nature Biotechnology 40 (8): 1231-40.

Piccinini, Filippo, Tamas Balassa, Abel Szkalisity, Csaba Molnar, Lassi Paavolainen, Kaisa Kujala, Krisztina Buzas, et al. 2017. “Advanced Cell Classifier: User- Friendly Machine-Leaming-Based Software for Discovering Phenotypes in High- Content Imaging Data.” Cell Systems 4 (6): 651-55. e5.

Smith, Kevin, Yunpeng Li, Filippo Piccinini, Gabor Csucs, Csaba Balazs, Alessandro Bevilacqua, and Peter Horvath. 2015. “CIDRE: An Illumination-Correction Method for Optical Microscopy.” Nature Methods 12 (5): 404-6.

Toth, Timea, Tamas Balassa, Norbert Bara, Ferenc Kovacs, Andras Kriston, Csaba Molnar, Lajos Haracska, Farkas Sukosd, and Peter Horvath. 2018. “Environmental Properties of Cells Improve Machine Learning-Based Phenotype Recognition Accuracy.” Scientific Reports 8 (1): 10085. Toth, Timea, Farkas Sukosd, Flora Kaptas, David Bauer, and Peter Horvath, n.d. “Show Me Your Neighbour and I Tell What You Are: Fisheye Transformation for Deep Learning-Based Single-Cell Phenotyping.” https://doi.org/10.1101/2022.08.23.505056.