Systems and methods for predictive molecular biomarker identification and quantification from morphology changes in histopathology tissue

1. A computer-implemented method of directly predicting molecular changes in one or more disease tissue sections from a subject, the method comprising:

analyzing molecular markers, which are predictive of tumor response and have no defined correlation with morphological features, such that the method is carried out in unsupervised manner; and

using artificial intelligence to correlate morphometric changes with critical gene modifications and molecular changes, wherein using the artificial intelligence comprises:

(i) collecting whole-slide images (WSIs) of tissue sections obtained from the subject;

(ii) feature extracting the whole-slide images into patches and corresponding vectors to generate clusters of vectors with morphologically similar patterns;

(iii) sampling the patches uniformly across each cluster to generate a batch of vectors that represent the subject's images;

(iv) generating a score between 0 and 1 for each patch based on gene status; and

(v) generating an outcome morphometric score for the subject by combining selected patches;

(vi) identifying regions of interest (ROIs) corresponding to morphological features correlated with disease outcome;

(vii) probe the ROIs with genes and protein specific probes;

(viii) identifying and quantifying RNA and proteins expressed in regions of interest (ROIs);

(ix) performing a correlation analysis between the RNA and the proteins expressed in the ROIs; and

(x) identifying biomarkers in the ROIs based on their expression or non-expression in other tissues, thereby identifying ROIs that are predictive of disease outcome.

2. The method of claim 1 , wherein an N×N region is selected around each patch, and a mean vector is generated for each region.

3. The method of claim 1 , wherein the molecular markers comprise PTEN, TMPRSS2-ERG, TP53, PIK3CA, MYC and ERBB2.

4. The method of claim 1 , wherein extracted features are trained on multiple tumor types, and wherein tumor types comprise prostate, breast, gastrointestinal, ovary, liver and lung tumors.

5. The method of claim 1 , wherein the tissue sections are stained with hematoxylin and eosin (H&E).

6. The method of claim 1 , wherein molecular changes comprise one or more gene modifications selected from gene mutations, copy-number variations (CNV), gene amplifications, gene deletions and gene fusions.

7. The method of claim 1 , wherein a score of 1 indicates a definite gene modification and a score of 0 indicates a wild type gene.

8. The method of claim 1 , wherein generating an outcome morphometric score for the subject indicates learning of predictive features.

9. The method of claim 1 , wherein the method predicts a range of gene modifications across a range of tumor types with 70 to 90% accuracy.

10. The method of claim 1 , wherein the method further comprises:

collecting patches causing miss-predictions and generating training data to fine-tune predictions; and

identifying regions of interest on tissue sections, which are predictive of outcome.

11. A system comprising one or more processors, and a non-transitory computer-readable medium including one or more sequences of instructions that, when executed by the one or more processors, cause the system to perform operations comprising:

analyzing molecular markers, which are predictive of tumor response and have no defined correlation with morphological features, such that the method is carried out in unsupervised manner; and

using artificial intelligence to correlate morphometric changes with critical gene modifications and molecular changes, wherein using artificial intelligence comprises:

(i) collecting whole-slide images (WSIs) of tissue sections obtained from the subject;

(ii) feature extracting the whole-slide images into patches and corresponding vectors to generate clusters of vectors with morphologically similar patterns;

(iii) sampling the patches uniformly across each cluster to generate a batch of vectors that represent the subject's images;

(iv) generating a score between 0 and 1 for each patch based on gene status; and

(v) generating an outcome morphometric score for the subject by combining selected patches;

(vi) identifying regions of interest (ROIs) corresponding to morphological features correlated with disease outcome;

(vii) probe the ROIs with genes and protein specific probes;

(viii) identifying and quantifying RNA and proteins expressed in regions of interest (ROIs);

(ix) performing a correlation analysis between the RNA and the proteins expressed in the ROIs; and

(x) identifying biomarkers in the ROIs based on their expression or non-expression in other tissues, thereby identifying ROIs that are predictive of disease outcome.

12. The system of claim 11 , wherein an N×N region is selected around each patch, and a mean vector is generated for each region.

13. The system of claim 11 , wherein the molecular markers comprise PTEN, TMPRSS2-ERG, TP53, PIK3CA, MYC and ERBB2.

14. The system of claim 11 , wherein extracted features are trained on multiple tumor types, and wherein tumor types comprise prostate, breast, gastrointestinal, ovary, liver and lung tumors.

15. The system of claim 11 , wherein the tissue sections are stained with hematoxylin and eosin (H&E).

16. The system of claim 11 , wherein molecular changes comprise one or more gene modifications selected from gene mutations, copy-number variations (CNV), gene amplifications, gene deletions and gene fusions.

17. The system of claim 11 , wherein a score of 1 indicates a definite gene modification and a score of 0 indicates a wild type gene.

18. The system of claim 11 , wherein generating an outcome morphometric score for the subject indicates learning of predictive features.

19. The system of claim 11 , further comprising the operations of:

predicting a range of gene modifications across a range of tumor types with 70 to 90% accuracy.

20. The system of claim 11 , further comprising the operations of:

collecting patches causing miss-predictions and generating training data to fine-tune predictions; and

identifying regions of interest on tissue sections, which are predictive of outcome.

21. The system of claim 11 , further comprising the operations of:

identifying and collecting patches from tissue sections for predicting the ROIs of tissue sections responsible for tumor progression;

performing univariate analysis using a paired t-test and the Benjamini-Hochberg method to control the false discovery rate (FDR) at <0.05 for identifying multiple univariate markers from the ROIs responsible for tumor progression;

performing multivariate linear regression analysis using a Cox proportional regression modeling to evaluate the associations between the multiple markers and a time to tumor progression; and

identifying a sub-set of the multiple markers according to a probability of tumor progression.

22. The system of claim 21 , further comprising the operations of:

constructing an overall risk score for time to recurrence based on a regression coefficient for each of the univariate markers.