Machine Learning Classification and Structure-Functional Analysis of Cancer Mutations Reveal Unique Dynamic and Network Signatures of Driver Sites in Oncogenes and Tumor Suppressor Genes.

In this study, we developed two cancer-specific machine learning classifiers for prediction of driver mutations in cancer-associated genes that were validated on canonical data sets of functionally validated mutations and applied to a large cancer genomics data set. By examining sequence, structure, and ensemble-based integrated features, we have shown that evolutionary conservation scores play a critical role in classification of cancer drivers and provide the strongest signal in the machine learning prediction. Through extensive comparative analysis with structure-functional experiments and multicenter mutational calling data from Pan Cancer Atlas studies, we have demonstrated the robustness of our models and addressed the validity of computational predictions. To address the interpretability of cancer-specific classification models and obtain novel insights about molecular signatures of driver mutations, we have complemented machine learning predictions with structure-functional analysis of cancer driver mutations in several important oncogenes and tumor suppressor genes. By examining structural and dynamic signatures of known mutational hotspots and the predicted driver mutations, we have shown that the greater flexibility of specific functional regions targeted by driver mutations in oncogenes may facilitate activating conformational changes, while loss-of-function driver mutations in tumor suppressor genes can preferentially target structurally rigid positions that mediate allosteric communications in residue interaction networks and modulate protein binding interfaces. By revealing molecular signatures of cancer driver mutations, our results highlighted limitations of the binary driver/passenger classification, suggesting that functionally relevant cancer mutations may span a continuum spectrum of driverlike effects. Based on this analysis, we propose for experimental testing a group of novel potential driver mutations that can act by altering structure, global dynamics, and allosteric interaction networks in important cancer genes.

Joint Region and Nucleus Segmentation for Characterization of Tumor Infiltrating Lymphocytes in Breast Cancer.

Histologic assessment of stromal tumor infiltrating lymphocytes (sTIL) as a surrogate of the host immune response has been shown to be prognostic and potentially chemo-predictive in triple-negative and HER2-positive breast cancers. The current practice of manual assessment is prone to intra- and inter-observer variability. Furthermore, the inter-play of sTILs, tumor cells, other microenvironment mediators, their spatial relationships, quantity, and other image-based features have yet to be determined exhaustively and systemically. Towards analysis of these aspects, we developed a deep learning based method for joint region-level and nucleus-level segmentation and classification of breast cancer H&E tissue whole slide images. Our proposed method simultaneously identifies tumor, fibroblast, and lymphocyte nuclei, along with key histologic region compartments including tumor and stroma. We also show how the resultant segmentation masks can be combined with seeding approaches to yield accurate nucleus classifications. Furthermore, we outline a simple workflow for calibrating computational scores to human scores for consistency. The pipeline identifies key compartments with high accuracy (Dice= overall: 0.78, tumor: 0.83, and fibroblasts: 0.77). ROC AUC for nucleus classification is high at 0.89 (micro-average), 0.89 (lymphocytes), 0.90 (tumor), and 0.78 (fibroblasts). Spearman correlation between computational sTIL and pathologist consensus is high (R=0.73, p<0.001) and is higher than inter-pathologist correlation (R=0.66, p<0.001). Both manual and computational sTIL scores successfully stratify patients by clinical progression outcomes.