NUCLEAR MAGNETIC RESONANCE

FIELD OF THE INVENTION

The present invention relates to a non-destructive method for predicting the germinating profile of maize seeds, using low field Nuclear Magnetic Resonance. Furthermore, the invention relates to a method for improving the germination ability of a maize seed lot, and a method for predicting the germination ability of a maize seed lot.

BACKGROUND OF THE INVENTION

Estimation of the germination ability of a seed lot may be performed for a variety of reasons. Mainly, it may be applied to determine germination ability of a seed lot or distinguish and select seeds that are able to develop into normal seedling under favorable conditions.

Indeed, the germination ability of a commercial seed lot is critical. For maize, it should be above 90% for a seed lot to be commercialized. Currently, this value is measured by the official methods of the International Seed Testing Association (ISTA) protocols. The seeds are placed in moistened filter paper of a defined quality under controlled humidity and temperature conditions over a specific period of time. After this time, the seeds that have germinated are counted manually by trained expert people. The main disadvantages related to the use of such tests are their time, space and electrical energy consuming. It further needs manual handling and specific training and it is destructive.

As the use of an official protocol to test this critical parameter of germination ability is mandatory, a need exists for a quick and non-destructive test to have a preliminary rapid and precise information about seed lot quality in upstream process.

Low-field NMR relaxometry is a nondestructive method, which is used to investigate water contents, mobility and interactions in matrix systems, such as plant organs. Deconvolution of the multiexponential curves obtained from longitudinal (Tl) and transverse (T2) relaxations allow identifying groups and proportions of water protons that share environments, interactions or molecular exchange rates. NMR relaxation time measurements have been used to characterize the water status of certain plant seeds during germination. In all of the studies already performed, NMR relaxation time measurements have identified three components of the water proton system.

In particular, the germination ability of soybean and wheat seeds are linked to storage condition (humidity and temperature) and NMR can be used to follow or detect seed affected by unappropriate storage conditions.

Krishnan P. et al., (Seed Science Research (2004) 14, 355-362) indeed disclosed that there were distinct changes in water status between viable and non- viable soybean seeds. In dry seeds, there were only two components, bound and bulk water, as revealed by analysis of T2. On the contrary, a three component water proton system (bound, bulk and free water) was observed in both germinating and nonviable soybeans during Phase 1 of hydration. The bulk water of non viable seeds disappeared completely during Phase 2 of hydration, resulting in a two component water proton system. This study also provided evidence that physical reorganization of water is essential in germinating soybean seeds during hydration.

Poulinquen D. et al., have reviewed the use of NMR on seed characterization (Poulinquen D. et al., Utilisation de la RMN en analyse de semences, communication avec actes, colloque biologie et qualite des semences, Angers decembre 98, presses de l'universite d'Angers (France) p 35-42), leading to the conclusion that a best knowledge of molecular dynamics and physiological and pathological modification may help to the nondestructive evaluation of physical and germinative quality of seed.

In the reported studies, the authors have focused on the effect of storage conditions on the seed germination. Conditions used in these experiments are drastic and cannot be considered as normal seed storage conditions.

Furthermore, from one variety of plant to another, the structure and the composition of the seed vary so much (for example, the proportion embryo/albumen, or the composition in water, starch, etc.) that NMR experiments with one plant species could not automatically be extrapolated to other plant species.

Therefore, the need still exists for a rapid and non-destructive solution to predict the germinating profile of maize seeds and detect non-germinating maize seed in real seed storage conditions of humidity and temperature (from real harvest to final storage going through usual seed process).

To the best of the inventor's knowledge, the use of low field NMR parameters have never been described for the prediction of a germinating profile of a maize seed lot.

SUMMARY OF THE INVENTION

Accordingly, it is first disclosed a method for predicting the germinating profile of a maize seed, the method comprising:

a) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on said maize seed, and

b) predicting the germinating profile of the seed based on the measurement(s) of step (a), using an appropriate mathematical model.

In one specific embodiment of said method, said one or more low field NMR parameter(s) is/are selected among the low field NMR parameter(s) which predicts the germinating profile. More specifically, said one or more low field NMR parameter(s) is/are advantageously selected among those which present a correlation coefficient with the germination ability of the seed, comprised between [-1; -0.6] and [0.6; 1].

In specific embodiments that may be combined with the previous embodiments, said one or more low field NMR parameter(s) is/are selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and,

(viii) combinations thereof.

In further specific embodiment, said one or more low field NMR parameter(s) is/are selected among the following:

(i) T2,

(ii) T2(l),

(iii) FID/P, (iv) A(l)/P, and

(v) Ampl/P.

In other specific embodiments that may be combined with the previous embodiments, the measurement(s) of step a) may be further completed by measurement(s) selected among visible spectroscopy, X ray 2D and 3D, fluorescence, and multi and hyperspectral spectroscopy. In these embodiments, the measurements can be combined and used in step b) for the prediction of a single germinating profile, or the prediction of the germinating profile can be performed multiple times using each data from the different methods.

In another specific embodiment that may be combined with the previous embodiments, measurement(s) of Near Infrared spectroscopy is/are performed on the seed before or after the step a) and the germinating profile is predicted based on the combined measurements ofNMR parameter(s) and RS parameter(s).

In another specific embodiment that may be combined with the previous embodiments, the seeds are calibrated by size before the measurements.

It is also disclosed a method for improving the germination ability of a maize seed lot, comprising

(i) predicting the germinating profile for each seed of said seed lot according to the above- described method, and,

(ii) sorting the seeds by retaining in the seed lot, either, the seeds which are predicted as germinating seeds, or discarding in the seed lot, the seeds which are predicted as non- germinating seeds,

thereby improving the germination ability of the seed lot.

In another embodiment, the method for improving the germination ability of a maize seed lot, comprises

(i) predicting the germinating profile for each seed of the seed lot by implementing the above-described method for predicting the germinating profile of a maize seed, and,

(ii) sorting the seeds of the seed lot by

- retaining in the seed lot the seeds which are predicted as germinating seeds, and discarding from the seed lot the seeds which are predicted as non-germinating seeds, thereby generating a seed lot with high predicted germination ability and another seed lot with low predicted germination ability. Another aspect of the present disclosure relates to a method for predicting the germination ability of a maize seed lot, comprising the steps of (i) obtaining a random sample of seeds from said seed lot, (ii) applying the above-described predicting method to said random sample of seeds, thereby obtaining a predicted germination ability value, and (iii) assessing the germination ability of the seed lot based on the predicted germination ability values of the random sample of seeds.

In a specific embodiment, the seed lot with low predicted germination ability is further analyzed to determine its germination ability by other non-NMR technique.

In another specific embodiment that may be combined with the previous embodiment, said other non-NMR technique includes NIR spectroscopy analysis of each seed of the random sample.

Another aspect of the present disclosure relates to an apparatus for carrying out the methods as defined above.

Typically, the apparatus of the present disclosure may comprise:

a signal source for generating NMR signal sequence appropriate for Tl and T2

NMR measurements,

a support for holding a seed or a seed lot,

a detector for detecting at least the NMR signal sequence having interacted with the seed or seed lot,

a data processing device for determining one or more NMR parameter(s), and more preferably one or more NMR parameter(s) selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and,

(viii) combinations thereof.

For example, the apparatus comprises:

1) a seed feeding system (1),

2) optionally, an individualization of seed system (2), 3) an NMR device for detecting one or more NMR signal sequence(s) having interacted with an individualized seed or seed lot (3),

4) optionally, an NIRS device (4),

5) a data processing device for determining the NMR parameter(s) (6),

6) a separator for sorting the seeds depending on the detected NMR or NMR and NIRS signals (5).

DETAILED DESCRIPTION A first aspect of the present disclosure relates to a method for predicting the germinating profile of a maize seed, said method comprising:

a) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on said maize seed, and

b) predicting the germinating profile of the seed, based on the measurement(s) of step (a), using an appropriate mathematical model.

The term "germination" as used herein refers to the first stage in the development of a plant from a seed, in particular the time at which the radicle breaks through the seed coat. Germination of the seed in a laboratory test is defined as the emergence and development of a seedling to a stage where the aspect of its essential structure indicates whether or not it is able to develop further into a satisfactory plant under favorable conditions in the field.

Similarly, as used herein, the term "germinating profile" of a seed refers to the classification of a seed, whether or not, it is a germinating seed, i.e. it will germinate and develop further into a satisfactory plant under favorable conditions in the field, or a non- germinating seed, i.e. it will not be able to germinate or it will not develop further into a satisfactory plant under favorable conditions in the field, as defined by the ISTA rules.

By "predicting" a germinating profile, it is meant in the context of the present invention that the method enables to determine the probability for a seed to have a certain germinating profile. More specifically, the method may allow to predict whether a seed is germinating or non-germinating as defined by the ISTA rules. The prediction is therefore not certain but associated to a specific probability, based on the mathematical method and the chosen threshold. It has been surprisingly found out that the measurement of a number of low field NMR parameters correlates with high significance with the germinating profile of a seed and/or the germination ability of a seed lot.

The method therefore provides one essential step of measuring one or more of such low field NMR parameter(s) on a maize seed, especially 'lI-NMR parameters.

Preferably, the one or more low field NMR parameter(s) for use in the above defined method is/are selected among the low field NMR parameters which can correlate with the germinating or non-germinating profile, as determined by the ISTA protocols, with a correlation coefficient comprised between [-1; -0.6] and [0.6; 1].

In one specific embodiment, said one or more low field NMR parameter(s) are selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and,

(xiiiv) combinations thereof.

In a specific embodiment, said one or more low-field NMR parameter(s) are selected among the following:

(i) T2,

(ii) T2(l),

(iii) FID/P,

(iv) A(l)/P, and,

(v) Ampl/P.

The skilled person will know how to measure these NMR parameters, which measurement methods can be classical NMR sequences such as those described for example in Singh K. and B. Blumich. NMR spectroscopy with compact instruments. Trends in Analytical Chemistry. Jan. 1, 2016; or, B. Blumich "Essential NMR" for Scientists and Engineers, 2005, Springer Berlin Heidelberg publisher. The parameter T2 represents the spin-spin relaxation times or transversal relaxation time. The parameter T2(l) represents the spin-spin relaxation times of the protons having short relaxation time. The parameter T2(2) represents the spin-spin relaxation times of the protons having long relaxation time. The parameter FID (Free Induction Decay) represents the signal induced in the coil after pulse excitation. The parameter Ampl, represents the amplitude of T2. The parameter A(l) represents the amplitude of T2(l). The parameter A(2) represents the amplitude of T2(2). P represents the mass of the maize seed or of the sample of seeds.

Accordingly, in a specific embodiment, the step of measuring one or more low field NMR parameter(s) will include:

applying a radio frequency pulse to a maize seed or a maize seed lot in the presence of an external low field magnetic field, for example a 20MHz field,

measuring spin-spin transverse relaxation times of protons (T2) and/or longitudinal relaxation time (Tl) using an NMR spectrometer,

determining the respective low field NMR parameters as defined above.

The conditions for acquiring the Tl and/or T2 NMR measurement(s) or other NMR parameters such as FID (Free Induction Decay) from the maize seed can be standard.

In one specific embodiment, the T2 measurements are obtained using the sequence as defined by Carr-Purcell-Meiboom-Gil (Snarr J. E. M. et H. Van As, 1992, Biophys. J. 63: 1654-1658).

In another specific embodiment, the Tl, T2, FID and other NMR parameter measurements are obtained as described in detail in Examples 1 and 2.

The method of the present disclosure may then predict the germinating profile based on the obtained low field NMR parameter values, as obtained from a maize seed, using an appropriate mathematical model.

The skilled person will know how to select an appropriate mathematical model, depending on the experimental designs. For example, the model may be based on a binary feature (germinating or non germinating), which model can be calibrated with a calibration dataset (i.e. control seeds with known NMR parameters values and germinating profile).

In a specific embodiment, the mathematical model used is a statistical analysis model like: y = F(Xa +s)

wherein J is a binary response variable ("germinating profile"),

X is a matrix containing predictors variables: variables accounting for the experimental design, NMR, NIRS variables...

a is a vector of effects for those variables and _ε is a residual, and

F is the logistic function such that Fit) = -,—r

^S ' l + exp(-t)

Predictions y _hat from this model are probability of germinating (between 0 and 1).

According to a peculiar feature of the invention, a threshold _c is preferably set such that if <C for a given seed, it is predicted to germinate, otherwise it is predicted not to germinate. By default, this threshold is set to 0.5. It can be modified to increase the detection power of non-germinating seed, at the cost of a higher false positive rate.

Because predictors variables in X are highly dimensional and correlated, for optimal predictive power, penalized regression methods are preferable to estimate _a ■ A first computation was done with the R glmnet (generalized linear models) package, with a « regularized logistic regression » (Friedman J. and al, (2010)), with the parameter T2 and variables to adjust for the experimental designs and the binary character germinated / non germinated as a response variable with a calibration dataset (seeds with both predictors (NMR, NIRS...) and the germinating profile.

For implementation details of this preferred embodiment, support can be found notably in the section 3 of the following paper: Friedman, J., Hastie, T. and R. Tibshirani (

Regularization Paths for Generalized Linear Models via Coordinate Descent, Journal of Statistical Software, Vol. 33(1), 1-22 Feb 2010).

This mathematical model is e.g. a Lasso regression, which is an analysis method that performs both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model it produces. A key feature of the Lasso is that it performs automatic feature selection such that the obtained model can be very sparse (with few coefficients in _a different from 0).

In a noteworthy embodiment, the binary feature for the germinating profile is "germinating" or "non-germinating" seed in real seed storage conditions of humidity and temperature, preferably according to ISTA rules. In certain embodiments, it may be advantageous to combine the low-field NMR measurements with other non-destructive measurements to increase the sensitivity and/or specificity of the prediction, in particular for predicting non-germinating maize seed.

Accordingly, in one specific embodiment, the above prediction method further comprises a measurements of Near Infrared spectroscopy performed on the seed before or after the step of measurements of said one or more NMR parameter(s).

Accordingly, in one embodiment, it is disclosed a method for predicting the germinating profile of a maize seed, the method comprising:

a) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on said maize seed,

b) measuring one or more Near infra-red spectroscopy (NIRS) parameter(s) on said maize seed,

c) predicting the germinating profile of the seed, using an appropriate mathematical model combining the measurements of steps (a) and (b).

The order of steps (a) and (b) is not essential.

It is also possible to use visible spectroscopy in addition to NIRS and/or NMR.

Preferably, said Near infra-red spectroscopy parameter(s) correspond to the measurement obtained on a maize seed at one or more of the following wavelengths: 1422 nm, 1896 nm, 1898nm, 1176nm, 1184nm, 1188 nm, 1678 nm, 1940 nm, 1944 nm, and 1968 nm. In another preferable method, near infra-red wavelengths used for the measurement are those of the area 1 to 4 in the table 14, or wavelengths with the higher effect as shown in figures 7 and 9. The wavelengths with the higher effect as shown in figures 7 and 9 also include VIS.

The preferred NMR parameters have been described previously and can be used advantageously in such specific embodiment. In a specific embodiment of such method combining NIRS and NMR parameter measurements, said one or more low-field NMR parameter(s) is/are selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P, (vi) A(l)/P,

(vii) A(2)/P, and,

(viii) combinations thereof.

The combined use of NMR and NIRS has demonstrated better prediction, compared to NMR alone.

In another embodiment, the NMR measurements of step a) of any of the above methods are further completed by measurement(s) selected among visible spectroscopy, fluorescence, 2D & 3D Xray, multi and hyperspectral imaging etc... Obviously more than one of these techniques can be used in addition to NMR measurements.

In a specific embodiment, the method for predicting the germinating profile of a maize seed, comprises the following steps:

a) measuring one or more low field Nuclear Magnetic Resonance (NMR) and/or NIRS parameter(s) on said maize seed, and

b) predicting the germinating profile of the seed, based on the measurement(s) of step (a), using an appropriate mathematical model.

Predicting the germination ability of a maize seed lot

It is further disclosed a method for predicting the germination ability of a maize seed lot, the method comprising:

a) obtaining a random sample of a maize seed lot,

b) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on said random sample of said maize seed lot, and

c) predicting the germination ability of the maize seed lot, based on the measurements of step (b), using an appropriate mathematical model.

As used herein, the term "germination ability" of the maize seed lot means the germination level (percentage) of seed lot, under favorable conditions, according to the ISTA protocols.

This method for predicting the germination ability of a maize seed lot can represent a complement to official ISTA protocols, leading to reliable and faster results. For "predicting" a seed lot germination ability, a previous step of seed sampling from the seed lot is needed and defined by the ISTA rules.

The seed lot germination ability is usually an estimation rate predicted based on the real germination observed on the seed sample according to ISTA rules. In the present method according to the invention, the seed lot germination ability is predicted based on the predicted germination ability of the random sample.

For seed lots which are susceptible to present a germination ability just below the threshold value mandatory for being commercialized, this method can thus be performed before the official one, for predicting seed lot certification.

In such situation, the ISTA protocol may be further used for confirming of infirming the first result as ISTA protocol is internationally used and even compulsory for seed lot certification (finished product stage).

The size of the random sample will depend on the required accuracy of the results. It will generally include between 10 and 500 seeds, for example between 20 and 200 seeds. For example, the measurements may be performed on 25, 100 or 200 maize seeds.

In one specific method, the prediction of the germination ability is based on the measurements of low-field NMR parameters, optionally combined with NIRS parameters, for each seed of the random sample and the prediction of their germinating profile as described in the previous sections.

Alternatively, the prediction is based on the measurement(s) of low-field NMR parameter(s), on the maize seed lot as a whole. The method is the same as the one performed on an individualized seed except that the low field NMR measurement is obtained on the random sample of maize seed lot, as a whole (in one measurement step).

Preferably, the one or more low field NMR parameter(s) for use in the above defined method is/are selected among the low field NMR parameters which can correlate with the germination ability of a maize seed lot as determined by the ISTA protocols, with a correlation coefficient comprised between [-1; -0.6] and [0.6; 1].

In one specific embodiment, said one or more low- field NMR parameter(s) is/are selected among the following: (i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and

(xiiiv) combinations thereof.

In a specific embodiment, said low-field NMR parameter(s) is/are selected among the following:

(i) T2,

(ii) T2(l),

(iii) FID/P,

(iv) A(l)/P,

(v) Ampl/P.

Similar to the method for predicting the germinating profile of a seed, a suitable mathematical model is applied to obtain a germination ability value of the sample of the maize seed lot, which correlates with the low-field NMR parameter(s) (optionally combined with NIRS parameter(s)) obtained on said sample of maize seed lot. Such germination ability value is also an estimate of the maize seed lot, as obtained from a random sample. In one preferred embodiment, the seed lot with low predicted germination ability is further analyzed to determine its germination ability by other non-NMR technique.

Determination of germination ability can be also done at any step of the seed process, including, for example, immediately following accidental damage on seed, or on aged seed lots, or on stored seeds.

Improving the germination ability of a maize seed lot

The method of predicting the germinating profile of a maize seed, as described above, can be used advantageously, in a first aspect, to improve the germination ability of a maize seed lot. In such embodiment, a method for improving the germination ability of a maize seed lot, comprises

(i) predicting the germinating profile for each seed of a maize seed lot according to the above-defined method, and,

(ii) sorting the seeds by, either, retaining in the seed lot the seeds which are predicted as germinating seed, or, discarding in the seed lot the seeds which are predicted as non-germinating seed, thereby improving the germination ability of the seed lot.

The step (i) of the above method is performed as described in the previous sections. The sorting step (ii) may be manual, semi-automatic or automatic.

A specific apparatus for carrying out said method is described in the next sections.

As used herein, the term "improving the germination ability" means that the maize seed lot obtained after the method has a higher germination ability than the maize seed lot obtained before carrying out the method.

In a specific embodiment, the method is applied to a maize seed lot with an initial germination ability which is below 0.90 (90%). In a related specific embodiment of the previous embodiment, the method is adjusted so that the predicted germination ability is increased to a level at least higher than 0.90 (90%), preferably, at least higher than 0.95% (95%).

In a specific embodiment, the low field NMR parameter(s) is/are selected among either (i) the T2 parameter, (ii) a combination of T2 and FID parameters, or (iii) all the NMR parameters without T2 parameter.

In one specific embodiment, the method for improving the germination ability of a maize seed lot, comprises the following steps:

(i) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on each maize seed of said maize seed lot,

(ii) measuring one or more Near infra-red spectroscopy ( RS) parameter(s) on each maize seed of said maize seed lot,

(iii) predicting the germinating profile of each seed, using an appropriate mathematical model combining the measurements of step (i) and (ii), and (iv) sorting the seeds by, either, retaining in the seed lot the seeds which are predicted as germinating seed, or, discarding in the seed lot the seeds which are predicted as non-germinating seed,

wherein the final seed lot obtained after the sorting step has an improved germination ability.

In one specific embodiment, the method for improving the germination ability of a maize seed lot, comprises the following steps:

(i) measuring one or more low field Nuclear Magnetic Resonance (NMR) parameter(s) on each maize seed of said maize seed lot,

(ii) measuring one or more Near infra-red spectroscopy ( RS) parameter(s) on each maize seed of said maize seed lot,

(iii) predicting the germinating profile of each seed, using an appropriate mathematical model combining the measurements of step (i) and (ii), and

(iv) sorting the seeds by

o retaining in the seed lot the seeds which are predicted as germinating seeds, and,

o discarding from the seed lot the seeds which are predicted as non- germinating seeds, thereby generating a seed lot with high predicted germination ability and another seed lot with low predicted germination ability

wherein the final seed lot obtained after the sorting step has an improved germination ability.

In a specific embodiment, the mathematical model used is a statistical analysis model like: y = F(Xa +s)

wherein J is a binary response variable ("germinating profile"),

X is a matrix containing predictors variables: variables accounting for the experimental design, NMR, NIRS variables...

a is a vector of effects for those variables and _ε is a residual, and

F is the logistic function such that Fit) = -,—r

^S ' l + exp(-t)

Predictions y _hat from this model are probability of germinating (between 0 and 1). According to a peculiar feature of the invention, a threshold _c is preferably set such that if <C for a given seed, it is predicted to germinate, otherwise it is predicted not to germinate. By default, this threshold is set to 0,5. It can be modified to increase the detection power of non-germinating seed, at the cost of a higher false positive rate. Because predictors variables in X are highly dimensional and correlated, for optimal predictive power, penalized regression methods are preferable to estimate _a . A first computation was done with the R glmnet (generalized linear models) package, with a « regularized logistic regression » (Friedman J. and al, (2010)), with the parameter T2 and variables to adjust for the experimental designs and the binary character germinated / non germinated as a response variable with a calibration dataset (seeds with both predictors (NMR, RS...) and the germinating profile.

For implementation details, see the section 3 of the following paper: Friedman, J., Hastie, T. and R. Tibshirani ( Regularization Paths for Generalized Linear Models via Coordinate Descent, Journal of Statistical Software, Vol. 33(1), 1-22 Feb 2010).

This mathematical model is advantageously a Lasso regression, which is an analysis method that performs both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model it produces. A key feature of the Lasso is that it performs automatic feature selection such that the obtained model can be very sparse (with few coefficients in _a different from 0).

In the above method, the order of steps (i) and (ii) is not essential.

In a specific embodiment combining NIRS and low-field NMR measurements, one or more (preferably all) of the following wavelengths of NIRS are used: 1422 nm, 1896 nm, 1898nm, 1176nm, 1 184nm, 1188 nm, 1678 nm, 1940 nm, 1944 nm, 1968 nm. . In another preferable method, near infra-red wavelengths used for the measurement are those of the area 1 to 4 in the table 14, or wavelengths with the higher effect as shown in figure 7 and 9. The wavelengths with the higher effect as shown in figures 7 and 9 also include VIS. It is also possible to use visible spectroscopy in addition to NIRS and/or NMR.

The apparatus for carrying out the above methods of the invention

In yet another aspect, the invention relates also to an apparatus for carrying out the methods for predicting the germinating profile of a maize seed or for predicting the germination ability of a maize seed lot, as previously described.

For predicting the germinating profile of a seed lot, a device can be implemented on the seed process with a random deviation of seeds sample from a seed lot, this seed sample being or not subsampled and used for NMR measurements and germination prediction of the global seed lot.

In yet another aspect, the invention relates also to an apparatus for carrying out the method for improving the germination ability of a maize seed lot, as previously described.

The apparatus according to the methods of the present disclosure typically includes

a signal source for generating NMR signal sequence appropriate for Tl and/or T2 NMR measurements,

- an NMR device for detecting at least the NMR signal sequence having interacted with the seed or seed lot,

a data processing device for determining the NMR parameter(s), and more preferably those selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and,

(viii) combinations thereof.

In an embodiment, the apparatus according to the methods of the present disclosure includes

- a radio -frequency transmit coil for Tl and/or T2 NMR measurements,

a radio-frequency receive coil for receiving the NMR signals emitted by the seed or seed lot,

a data processing device configured to determine the NMR parameter(s), and more preferably those selected among the following:

(i) T2,

(ii) T2(l),

(iii) T2(2),

(iv) FID/P,

(v) Ampl/P,

(vi) A(l)/P,

(vii) A(2)/P, and,

(viii) combinations thereof. In another specific embodiment, the data processing device further includes a seed germinating profile output, providing for each seed, a result of its germinating profile, whether germinating or non-germinating.

In another specific embodiment, the apparatus may further include one or more of the following:

a seed feeding system, upstream of the apparatus, for feeding the seed, one by one, into the apparatus,

a separator, downstream of the apparatus, the separator having a control input connected to the seed germinating profile output, the separator being arranged to separate the seed predicted as germinating seed from the seed predicted as non- germinating seed.

Such apparatus may therefore include:

1) a seed feeding system (1),

2) optionally, an individualization of seed system (2),

3) an NMR device for detecting one or more NMR signal sequence having interacted with an individualized seed or seed lot (3),

4) optionally, an NIRS device (4),

5) a separator for sorting the seed depending on the detected NMR or NMR and NIRS signals (5).

Individualization can be achieved by an individualization system comprising a hopper supplied with corn kernels by a vibrating plate. The hopper drives the corn kernels in a system with two endless screws having an increasing pitch. These screws enable speed of the corn kernels to be increased and corn kernels to be separated from each other so as to be individualized. Individualized kernels can laid onto the support surface of a conveyor and two guiding elements can position the corn kernel in the median part of the conveyor.

In an embodiment, such apparatus includes:

1) a seed feeding system (1),

2) optionally, an individualization system comprising a hopper supplied with corn kernels by a vibrating plate (2),

3) a radio-frequency receive coil for receiving the NMR signals emitted by the seed or seed lot (3),

4) optionally, an NIRS device (4), 5) a separator for sorting the seed depending on the detected NMR or NMR and NIRS signals (5).

A NIRS device suitable for measuring NIRS parameters from seed is commercialized by Polytec (NIR spectrometer on diode array technology), but also by Zeiss or Specim. More than one spectrometer can be used simultaneously to cover the NIR spectra.

Preferably, upstream of the NIRS device, the apparatus may include a system for orienting the seed with respect to its embryo for NIRS measurement. The system for orienting the seed may comprise at least one laser device arranged to lighten the corn kernel with a laser line, and a plurality of orientation imaging devices configured to acquire respective two- dimension orientation images of the corn kernel along different viewing directions. During the step of determining an orientation of the corn kernel, the orientation of the corn kernel with respect to the support surface may then be determined based on the structural features of the corn kernel measured on each of the two-dimension orientation images.

A data processing device (6) may be connected to the NMR device, the NIRS device and the separator.

An embodiment of such apparatus is represented in Figure 6. Such specific embodiment of apparatus is useful in the method for improving the germination ability of a maize seed lot.

Method for predicting the germinating profile of a maize seed using NIRS

It has been shown that NIRS values also correlate with the germinating profile of a maize seed. Such NIRS values may be used alone or in combination with NMR parameter values in methods for predicting the germinating profile of a maize seed using NIRS.

In another aspect, it is disclosed herein a method for predicting the germinating profile of a maize seed, the method comprising:

a) measuring near infra-red spectroscopy values on said maize seed, and

b) predicting the germinating profile of the seed based on the measurement of step (a), using an appropriate mathematical model.

In a specific embodiment, the mathematical model used is a statistical analysis model like: y = F(Xa +s)

wherein J is a binary response variable ("germinating profile"), X is a matrix containing predictors variables: variables accounting for the experimental design, NMR, NIRS variables...

a is a vector of effects for those variables and _ε is a residual, and

F is the logistic function such that Fit) = -,—r

^S ' l + exp(-t)

Predictions y _hat from this model are probability of germinating (between 0 and 1). According to a peculiar feature of the invention, a threshold _c is preferably set such that if <C for a given seed, it is predicted to germinate, otherwise it is predicted not to germinate. By default, this threshold is set to 0,5. It can be modified to increase the detection power of non-germinating seed, at the cost of a higher false positive rate.

Because predictors variables in X are highly dimensional and correlated, for optimal predictive power, penalized regression methods are preferable to estimate _a ■ A first computation was done with the R glmnet (generalized linear models) package, with a « regularized logistic regression » (Friedman J. and al, (2010)), with the parameter T2 and variables to adjust for the experimental designs and the binary character germinated / non germinated as a response variable with a calibration dataset (seeds with both predictors (NMR, NIRS...) and the germinating profile.

For implementation details, see the section 3 of the following paper: Friedman, J., Hastie, T. and R. Tibshirani ( Regularization Paths for Generalized Linear Models via Coordinate Descent, Journal of Statistical Software, Vol. 33(1), 1-22 Feb 2010).

This mathematical model is advantageously a Lasso regression, which is an analysis method that performs both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model it produces. A key feature of the Lasso is that it performs automatic feature selection such that the obtained model can be very sparse (with few coefficients in _a different from 0).

In a specific embodiment, at step a), NIRS values are measured at one or more of the following wave-lengths: 1422 nm, 1896 nm, 1898nm, 1176nm, 1184nm, 1188 nm, 1678 nm, 1940 nm, 1944 nm, 1968 nm. In another preferable method, near infra-red wavelengths used for the measurement are those of the area 1 to 4 in the table 14, or wavelengths with the higher effect as shown in figure 7 and 9. The wavelengths with the higher effect as shown in figures 7 and 9 also include VIS. Similar to the method for predicting the germinating profile of a seed using NMR parameters, a suitable mathematical model is applied to determine whether a seed is germinating or non-germinating based on the NIRS values.

It is also possible to use visible spectroscopy in addition to NIRS and/or NMR.

Specific embodiments of the invention will now be illustrated in the following examples.

It is further disclosed a method for improving the germination ability of a maize seed lot, comprises the following steps:

(i) measuring one or more Near infra-red spectroscopy (NIRS) parameter(s) on each maize seed of said maize seed lot,

(ii) predicting the germinating profile of each seed, using an appropriate mathematical model, and

(iii) sorting the seeds by

o retaining in the seed lot the seeds which are predicted as germinating seeds, and,

o discarding from the seed lot the seeds which are predicted as non- germinating seeds, thereby generating another seed lot with low predicted germination ability

wherein the final seed lot obtained after the sorting step has an improved germination ability.

In yet another aspect, the invention relates also to an apparatus for carrying out the method for improving the germination ability of a maize seed lot, as described above. Such apparatus may include:

1) a seed feeding system,

2) optionally, a vibrating table,

3) a NIRS device,

4) a separator for sorting the seed depending on the detected NIRS signals. LEGENDS OF THE FIGURES:

Figure 1: Correlation between the germination ability (% germination) and the low- field NMR parameters, for the same seed lot at a different step in the seed process: (a) T2, (b) T2(l), (c) FID/P, (d) Ampl/P and (e) A(l)/P. Figure 2: Correlation between the germination ability (% germination) and the low- field NMR parameters for different seed varieties: (a) FID/P, (b) A(l)/P.

Figure 3: illustrates for each wavelength the correlation with the germination ability. Relevant wavelengths for the model are those with a correlation coefficient different from zero.

Figure 4: Comparison of germination ability of seed lots after sorting seed according to the germinating profile predicted on different NMR parameters compared to germination ability before (without sorting).

Figure 5: Comparison of germination ability of seed lots after sorting seed according to the germination profile predicted on all NMR parameters, with the same mathematical model and a different threshold compared to germination ability before (without sorting).

Figure 6: Embodiment of an apparatus according to the invention combining NMR and NIRS parameters detection.

Figure 7: This figure illustrates for each wavelength the relevance to predict the germination ability, relevant wavelengths for the model are those with a value (effect) different from zero

Figure 8: This figure illustrates for each wavelength the correlation between repeated measures

Figure 9: This figure illustrates for each wavelength the relevance to predict the germination ability, the thousand kernel weight (TKW) has been include in the model, relevant wavelengths for the model are those with a value (effect) different from zero.

Figure 10: This figure illustrates for each parameter (NMR and NIRS and VIS spectroscopy) used in example 7, the relevance to predict the germination ability, relevant parameters are those with a value (effect) different from zero. In this figure "signal average" means FID.

EXAMPLES

Example 1: Identification of an indicator of the germination ability The experiment was conducted on the same seed lot of maize.

Five independent samples of this lot were extracted at different seed process stages.

Lot A correspond to freshly shelled stage,

Lot B corresponds to filling truck after having been shelled

and Lot C is truck sample on its arrival on new factory (after transfer).

The germination ability was measured by a conventional protocol on these lots, according to ISTA rules, the average values of the 5 lots are respectively 96,6% for lot A, 92,4% for lot B and 90,6% for lot C.

Low- field NMR measurements were taken on five samples of 5 kernels for each one of the three stages A, B and C.

More specifically, the low-field NMR measurements were carried out on a Minispec mq 20 of the Bruker type running at 20MHz.

Spin-spin relaxation times (T2) or transversal relaxation time and corresponding amplitude (Ampl) were measured by the Carr-Purcell-Meiboom-Gil sequence (Snarr J. E. M. and H. Van As, 1992). The acquisition conditions are as follows: number of points (150), pulse separation (0.5 ms), dummy echo (3) and number of scans (10). These experimental conditions are identical for each analysis. The gain was adjusted so as to maximise the signal-to-noise ratio (Snarr, J.E.M. and H. Van As. 1992. Probing water compartments and membrane permeability in plant cells by H NMR relaxation measurements. Biophys. J. 63: 1654-1658).

The bi-exponential relaxation decay curves are measured in the same experimental conditions than T2 to discriminate between the different protons families (A(l) and A(2)) having respectively short or long relaxation time (T2(l) and T2(2)).

The amplitudes and relaxation times of the different components were extracted from bi- exponential decay curves using the apparatus software. The FID corresponds to the signal intensity recovered after a 90° excitation pulse. P is the weight of the sample (g), FID/P, Ampl/P, A(l)/P and A(2)/P are respectively FID, Ampl, A(l) and A(2) value divided by the sample weight. As shown in Figure 1, a clear correlation between germination ability was revealed for the following NMR parameters:

(i) T2,

(ii) T2(l),

(iii) FID/P,

(iv) Ampl/P, and,

(v) A(l)/P.

This example shows the possibility of sampling kernel seed lot and of predicting the germination ability of the seed lot based on certain low- field NMR parameters and interpretation on such sample.

Example 2:

The experiment was conducted on 11 lots of 384 maize kernels each. Each lot corresponds to kernels from different varieties, and these lots are at bulk shelled level i.e no sized.

The following 7 low-field NMR parameters were used:

Table 1: low- fie Id NMR parameters measured in the experiment.

The low- field NMR analyses, carried out on a Minispec mq 20 of the Bruker type running at 20MHz, were conducted on each grain.

Spin-spin relaxation time (T2) or transversal relaxation time and corresponding amplitude (Ampl) were measured by the Carr-Purcell-Meiboom-Gil sequence (Snarr J. E. M. and H. Van As, 1992). The acquisition conditions are as follows: number of points (150), pulse separation (0.5 ms), dummy echo (3) and number of scans (10). These experimental conditions are identical for each analysis. The gain was adjusted so as to maximize the signal-to-noise ratio (Snarr, J.E.M. and H. Van As. 1992. Probing water compartments and membrane permeability in plant cells by H NMR relaxation measurements. Biophys. J. 63: 1654-1658).

The bi-exponential relaxation decay curves are measured in the same experimental conditions than T2 to discriminate between the different protons families (A(l) and A(2)) having respectively short or long relaxation time (T2(l) and T2(2)).

The amplitudes and relaxation times of the different components were extracted from bi- exponential decay curves using the apparatus software. The FID corresponds to the signal intensity recovered after a 90° excitation pulse.

T2, T2(l) and T2 (2) are expressed in ms, FID, Ampl, A(l) and A(2) in AU arbitrary unit, and expressed per g of sample.

At the end of the analytical phase, all of maize kernels were placed in germination test (germination 25°C, 7 days, sand, 14% H ₂ 0), seed can then be sorted into germination or non-germinating classes and each seedling coming from each seed was able to be characterized and sorted as normal or abnormal seedling according to its germination defects.

For each batch, a germination ability is calculated by taking an average of the values attributed for the 384 seeds which is then expressed as a percentage. This percentage is referred as the calculated germination ability (FGcal).

For each NMR parameter, an average is taken for each variety and this value is listed across from the variety. A correlation is sought between the NMR parameter and FGcal: The correlation coefficient is determined in order to know if there is a link between the 2 variables and how strong such link is. The closer the correlation coefficient is to 1 or -1, the stronger the link is between the 2 variables.

A correlation is shown for the following two variables:

Table 2: Correlation coefficient between germination ability and NMR parameters

Figure 2 further illustrates the correlation between germination ability of ten maize seed lots and the obtained NMR parameters values. This correlation makes it possible to classify the lots based on the NMR parameters values. For example, a lot that obtains an NMR value for FID/P greater than 210 UA/g, is assumed to have a germination ability greater than 90% and therefore can be marketed in accordance with ISTA rules.

A lot that has a value in the neighborhood of 210 UA/g or below will potentially have a germination ability defect and treatment for it in the factory will have to be adapted so as to meet this standard.

By reproducing this correlation analysis, the skilled person will be able to determine appropriate threshold values, depending on the specific requirements for the lots to be sorted.

EXAMPLE 3: Improvement of the germination ability of a maize seed lot a- Mathematical model

The statistical analysis model used is:

y = F(Xa +s)

wherein J is a binary response variable ("germinating profile"),

X is a matrix containing predictors variables: variables accounting for the experimental design, NMR, NIRS variables...

a is a vector of effects for those variables and _ε is a residual, and

F is the logistic function such that Fit) = -,—r

^S ' l + exp(-t)

Predictions y _hat from this model are probability of germinating (between 0 and 1). A threshold _c has to be set such that if y _M <C for a given seed, it is predicted to germinate, otherwise it is predicted not to germinate. By default, this threshold is set to 0,5. It can be modified to increase the detection power of non-germinating seed, at the cost of a higher false positive rate.

Because predictors variables in X are highly dimensional and correlated, for optimal predictive power, penalized regression methods are needed to estimate _a ■ A first computation was done with the R glmnet (generalized linear models) package, with a « regularized logistic regression » (Friedman J. and al, (2010)), with the parameter T2 and variables to adjust for the experimental designs and the binary character germinated / non germinated as a response variable with a calibration dataset (seeds with both predictors (NMR, RS...) and the germinating profile.

For implementation details, see the section 3 of the following paper: Friedman, J., Hastie, T. and R. Tibshirani ( Regularization Paths for Generalized Linear Models via Coordinate Descent, Journal of Statistical Software, Vol. 33(1), 1-22 Feb 2010).

This model is a Lasso regression, which is an analysis method that performs both variable selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model it produces. A key feature of the Lasso is that it performs automatic feature selection such that the obtained model can be very sparse (with few coefficients in a different from 0).

This model was used for cross validation with the set of data to predict the germinating profile of one lot. By default the threshold value is equal to 0,5.

The results are presented in Tables 3 to 7. Table 3 corresponds to the sorting based on T2 NMR parameter. Table 4 corresponds to the sorting based on T2 & FID parameters. Table 5 is the sorting based on all NMR parameters except T2. Finally Tables 6 and 7 corresponds to the sorting made with all NMR parameters but with different threshold (see also Figures 4 and 5).

As used in the tables below:

"Lot" is the reference of the lot,

"% identified" is the ratio of seed identified as non-germinating seeds,

"% false positive" is the number of germinating seed identified as non-germinating seed, "germination rate before" is the germination rate of the lot without sorting, "germination rate after" if the germination rate after elimination of the seed identified as non- germinating, and,

"≠ case" is the effective number of non-germinating seed in the lot. b- Sorting of non-germinating seeds on T2 parameter.

Table 3: prediction of germinating vs. non-germinating seeds with the model according to the invention. The model is constructed on the T2 value (threshold 0,5).

For two lots with a germination ability better than 90%, few or non-germinating seeds have been identified, and impact of screening has been low on the germination rate. For the others lots, the rate of germinating seeds has been increased until 5 %. This shows the efficiency of the method to increase germination ability of a seed lot.

However germination ability for some lots is still under 90%, for this sorting by a threshold of 0.5, this threshold should be adapted to discard more seeds and still improve the germination ability of the lot.

c- Sorting of non-germinating seeds on T2 and FID parameters

Table 4: Prediction of germinating vs. non-germinating seeds with the model of the invention. The model is constructed on the T2 and FID values (threshold 0,5)

For two lots with a germination ability better than 90%, few or non-germinating seeds have been identified, and impact of screening has been low on the germination rate. For the other lots, the rate of germinating seeds has been increased until 6 %. This shows the efficiency of the method to increase germination ability of a seed lot.

However germination ability for some lots is still under 90%, for this sorting by a threshold of 0.5, this threshold should be adapted to discard more seeds and still improve the germination ability of the lot. d- Sorting of non-germinating seeds on all parameters without T2.

Table 5: Prediction of germinating vs. non-germinating seeds with the model of the invention. The model is constructed with all the values without T2 (threshold 0,5).

For two lots with a germination ability better than 90%, few or non-germinating seeds have been identified, and impact of screening has been low on the germination rate. For the other lots, the rate of germinating seeds has been increased until 1 %. This shows the efficiency of the method to increase germination ability of a seed lot.

However germination ability for some lots is still under 90%, for this sorting by a threshold of 0.5, this threshold should be adapted to discard more seeds and still improve the germination ability of the lot. e- Sorting of non-germinating seeds with all the NMR parameters Further computation was done, based on the use of all the NMR values (Table 6).

Table 6: Prediction of germinating vs. non-germinating seeds with the model of the invention including all the NMR values (threshold 0.5).

For two lots with a germination ability better than 90%, few or non-germinating seeds have been identified, and impact of screening has been low on the germination rate. For the other lots, the rate of germinating seeds has been increased until 5 %. This shows the efficiency of the method to increase germination ability of a seed lot.

However germination ability for some lots is still under 90%, for this sorting by a threshold of 0.5, this threshold should be adapted to discard more seeds and still improve the germination ability of the lot. The following tables show how the % germination rate change when the threshold is decreased. The same calculation was done with a threshold of 0,3 (Table 7).

Table 7: Prediction of germinating vs. non-germinating seeds with the model of the invention including all the NMR values (threshold (0,3).

The rate of germinating seeds has been increased until 18 %. This shows the efficiency of the method to increase germination ability of a seed lot. The results show that 90% of germination or above can be reached when decreasing the threshold (see also Figure 5).

EXAMPLE 4: identification of germination ability by Near Infrared Spectroscopy (NIRS)

The Foss 6500 module is used for NIRS analysis with a mono seed ring cup. The seed is manually introduced in the ring cup with a proper orientation according to the embryo. Analysis is done for wavelength from 800 to 2500 nm with a step of 2nm, the analysis is of 1 min for each sample.

The same seeds lots were analyzed by NIRS. As shown in Figure 3, a correlation between NIRS values and germination ability has been observed using similar mathematical model for determining correlation as described in Example 3. 482 wavelengths presenting a correlation with the germination have been identified These results show than NIRS data can also be used to predict seed germination. In particular, the following wavelengths show a higher prediction ability: 1422 nm, 1896 nm, 1898nm, 1176nm, 1184nm, 1188 nm, 1678 nm, 1940 nm, 1944 nm, 1968 nm.

The same statistical model was used with NIRS data as the model used in Example 3. Results are presented in the following tables 8, 9, 10.

Table 8 describes the germination rate improvement based on NIRS data model combining all relevant 482 wavelengths.

Table 9 shows how that the same model can be improved by injecting 3 extra lots data. Finally Table 10 shows the germination rate improvement when using only 10 out of 482 wavelengths (chosen amongst best ones). Legends of the tables are identical to those of example 3.

Table 8: Prediction of germinating vs non-germinating seeds with a model including all NIR spectra for the 482 wavelength according to the invention. The model is constructed on whole wavelength values, these values are regularized by a lasso type penalization and a threshold of 0.5.

Germination ability has been improved for 6 of the 11 seeds lots. This analysis has been completed by measurements performed on 3 seed lots with a very low germination ability (Table 9).

Table 9: Prediction of germinating vs non-germinating seeds with a model including all NIR spectra for the 482 wavelength to the invention, the model is constructed on all wavelength values, these values are regularized by a lasso type penalization and a threshold of 0.5.

The model is improved (probably by the asset of data more important in number and in diversity) and the ability of the method to enable germination rate improvement of lots is confirmed: 8 of the 11 lots have been improved and the mean of improvement reaches 9,75%.

The following table 10 shows the improvement of germination ability using the following 10 best wave-lengths 1422 nm, 1896 nm, 1898nm, 1176nm, 1184nm, 1188 nm, 1678 nm, 1940 nm, 1944 nm, 1968 nm. 4 lots have been improved. NIR spectra 10 best wavelengths

% % false

Lot identified positive germ rate before germ rate after # cases

VO 0 0 0.98 0.98 8

VI 3.57 0 0.93 0.93 28

V2 0 0 0.85 0.86 55

V3 2.33 0 0.77 0.78 86

V4 0 0 0.80 0.80 77

V5 0 0 0.87 0.87 49

V6 0 0 0.87 0.87 48

V7 0 0 0.86 0.86 52

V8 4.26 0 0.75 0.77 94

V9 0 0 0.66 0.66 130

V10 1.67 0 0.84 0.85 60

Table 10: prediction of germinating and non-germinating seeds with the model of the invention, the model being constructed on the 10 wavelength with the best correlation with germination ability (fig. 3). The model is constructed on 10 wavelength values, these values are regularized by a lasso type penalization and a threshold of 0.5.

EXAMPLE 5: combination of the use of NMR and NIRS for predicting and improving germination ability of a seed lot.

In the following protocol, a first sorting is done on NMR values with the method as described in Example 3, intermediate germination rate is the rate of the seed after this first sorting. A second sorting is done on remaining seeds by the NIRS method as described in the example 4 using all the wavelength measurements. The germination rate after is the final germination rate. The results are shown in Table 11. In the table below, "# cases" refers to the number of non-germinating seeds.

Table 11: Improvement of the germination ability of a seed lot comprising, prediction of germinating vs. non-germinating seeds with the model of the invention on NMR values and discarding non germinating seed followed by prediction on remaining seed by the model of the invention on NIRS values and discarding non germinating seeds (0.5 threshold used for both predictions).

In the next protocol, a first sorting is done on NIRS values, intermediate germination rate is the rate of the seed after this first sorting. A second sorting is done on remaining seeds by the NMR method and the germination rate after is the final germination rate. The results are shown in Table 12.

As used in the table below, "# cases" number of non-germinating seeds.

Table 12: Improvement of the germination ability of a seed lot comprising, prediction of germinating vs. non-germinating seeds with the model of the invention on NIRS values and discarding non germinating seed followed by prediction on remaining seed by the model of the invention on NMR values and discarding non germinating seeds. 0.5 threshold used for both predictions NMR

In the next protocol, NMR and NIRS measurements are done and the sorting is done according to the method on an estimation being based on the combination of both results. "# cases" number of non-germinating seeds.

% % false germination rate germination rate

Lot identified positive before (%) after (%) # cases

VO 37.5 9.84 0.98 0.99 8

VI 7.14 0.84 0.93 0.95 28

V2 3.64 1.22 0.85 0.86 55

V3 6.98 0.67 0.77 0.80 86

V4 9.09 1.63 0.80 0.83 77

V5 6.12 0 0.87 0.88 49

V6 0 0.30 0.87 0.87 48

V7 0 0 0.86 0.86 52

V8 42.55 0.69 0.75 0.95 94

V9 10.77 1.18 0.66 0.71 130

V10 16.67 0.312 0.84 0.89 60

Table 13: NIRS and NMR measurements are done sequentially, but the model includes correlation and sorting on both parameters. 0,5 threshold used. The results of Table 13 shows that, if the sorting is based on a model combining the values of NIRS and NMR measurements, the germination ability and prediction is further improved.

EXAMPLE 6: identification of germination ability by NIR (Near Infrared Spectroscopy) and VIS (visible) Spectroscopy.

The Foss 6500 module offers a larger range of spectra from 400 nm to 2500 nm this spectra include visible light and NIR.

This experiment was conducted on 45 lots of 336 maize kernel each i.e more than 15000 kernels. These lots are from 27 different varieties, at shelled level i.e no sized covering the usual germination range. These lots have been analyzed 2 to 10 weeks after the harvest With the protocol of example 4 correlation between NIR and VIS values and germination ability have been observed using mathematical model for determining correlation as described in example 3. Results are illustrated in Figure 7.

The most relevant wavelength identified are listed in Table 14. These wavelength define some area of interest on the spectra scanned by the experiment, area 2 and 3 are common to the experiment of example 4. The experiment 6 performed on seed analyzed 2 to 10 weeks from the harvest confirm results of example 4 done on seed 1 year on average after harvest. This experiment 6 shows a new zone in the lower part of the NIR spectrum (700- 2500 nm) at the neighborhood of the visible spectra.

Table 14: the most relevant wavelength for predicting germination ability obtained for the example 5, line 1 and example 4, line 2. The third line are the zones defined as relevant by the experiment (1 to 4) a) repeatability of relevant wavelength

Repeatability has been tested with a plate of 48 seeds from VI, these seeds have been tested twice at the beginning and at the end of the measurement of the whole experiment. The correlation between repeated measures for each wavelength are shown in figure 8. This experiment shows that measures for the same seeds can be repeatable after a delay of 15 weeks. NIR answer is stable along the sample storage after harvest. For the future model selected predictive wavelength should be chosen between predictive and repeatable wavelength. b) TKW (Thousand kernel weight)

The weight of the seed has been introduced in the model, and the correlation has been slightly improved (figure 9). This result indicate that the germination prediction by the model should be improved if used on sized seeds lots. c) Sorting of non-germinating seeds on NIRS parameters

The same cross validation statistical model was used with NIRS data as the model used in Example 3. Results are presented in the following table 15 which describes the germination rate improvement based on NIRS data model using all wavelength in a penalized logistic regression model. Lot % identified % false germination rate germination rate # cases positive before (%) after (%) initial

VI 14,29 0,54 0,96 0,97 7

V2 25 1,54 0,96 0,98 12

V3 28,57 1,22 0,98 0,99 7

V4 6,45 3,28 0,91 0,91 31

V5 7,41 1,62 0,92 0,93 27

V6 34,29 4,98 0,89 0,96 35

V7 11,76 2,32 0,90 0,92 34

V8 6,67 0,62 0,96 0,96 15

V9 16 0,96 0,93 0,95 25

V10 0 1,83 0,98 0,98 8

Vl l 11,11 1,83 0,97 0,98 9

V12 16,67 4,40 0,95 0,96 18

V13 42,86 1,22 0,98 1 7

V14 9,53 0,32 0,94 0,95 21

V15 4,35 0,32 0,93 0,94 23

V16 13,04 0 0,93 0,95 23

V17 0 0,30 0,99 0,99 5

V18 9,09 0,92 0,97 0,97 11

V19 50 1,85 0,96 1 12

V20 11,11 0,31 0,97 0,98 9

V21 0 0 0,98 0,98 6

V22 13,33 1,92 0,95 0,96 15

V23 0 0,61 0,98 0,98 6

V24 14,28 0,61 0,98 0,98 7

V26 0 0,91 0,98 0,99 4

V27 40 0,92 0,97 0,99 10

V28 25,57 1,55 0,96 0,98 14

V29 20 1,53 0,97 0,98 10

V30 0 0,61 0,97 0,97 9

Table 15: Prediction of germinating vs non-germinating seeds with a model using NIRS data. The model is constructed on whole wavelength values, these values are regularized by a lasso type penalization and a threshold of 0.2.

Germination ability has been improved for 20 of the 29 seeds lots. EXAMPLE 7: combination of the use of NIR and VIS spectroscopy and NMR for predicting and improving germination ability of a seed lot,

In the following protocol, a first sorting is done on NIR and VIS spectroscopy values with the method as described in Example 6 (all the wavelength measurements), intermediate germination rate is the rate of the seed after this first sorting.

A second sorting is done on remaining seeds by the NMR method as described in the example 4. The germination rate after is the final germination rate, the results are shown in Table 16.

In the table below, "# cases" refers to the number of non- erminatin seeds,

Table 16: Improvement of the germination ability of a seed lot comprising, prediction of germinating vs. non-germinating seeds with the model of the invention on NIR and VIS values and discarding non germinating seed followed by prediction on remaining seed by the model of the invention on NMR values and discarding non germinating seeds (0.5 threshold used for both predictions).

The figure 10 illustrates for each parameter (NMR and NIRS and VIS spectroscopy) used in the example, the relevance to predict the germination ability, relevant parameters are those with a value (effect) different from zero. In this figure "signal average" means FID.