PhD Proposal: Computational Techniques for the Analysis of Inter and Intra-tumor Heterogeneity
Cancer is a heterogeneous disease. This heterogeneity can be of two types. (1) Inter tumor heterogeneity, which captures variability of the disease among different individuals and (2) Intratumor heterogeneity, which captures variability within the same individual. With this in mind, we set out to build computational analysis techniques to further our understanding of inter and intra tumor heterogeneity.First, we set out to explore the extent of inter tumor functional heterogeneity in breast cancers. To mathematically model inter-tumor functional heterogeneity, we exploited the concept of guilt by association, which essentially says that a protein’s function depends on its Protein-Protein Interaction (PPI) neighborhood. Thus, a protein may gain or lose functions if its PPI neighborhood changes. Our analysis of 1047 breast cancer samples from The Cancer Genome Atlas (TCGA) reveal that breast cancers possess a high degree of inter-tumor functional heterogeneity that cannot be simply explained by gene expression differences. Furthermore, these predicted changes in protein function are associated with an elevated missense mutation frequency and patient survival.Second, we set out explore the relationship between intra tumor genetic heterogeneity and immune surveillance in skin cancer. We analyzed Whole Exome Sequencing (WES), SNP array and gene expression data of 330 melanomas from TCGA to find that high intra-tumor genetic heterogeneity is associated with low immune cytotoxicity and worse patient survival, independent of other clinical factors. Furthermore, for melanomas, intra tumor genetic heterogeneity was a better predictor of response to immunotherapy than mutation load. To verify our results, we are conducting experiments that monitor the immune response of mice to melanomas with varying degree of intra-tumor genetic heterogeneity and mutation load.Third, we set out to study the chromosome level copy number heterogeneity across cancers in different tissues. Previous studies have shown that the copy number of chromosome arms in cancer is tissue specific. However, it has been challenging to find out why. We discover a striking similarity between the tissue specific copy number of a chromosome arm in cancer and its tissue specific gene expression in healthy tissue. We thus hypothesize that these tissue specific copy number alterations are physiological constraints placed by the tissue on genomic instability in early stages of cancer evolution. As a follow-up, we plan to test whether the tissuespecific copy number of chromosome arms in cancer is associated with their tissue specific DNA methylation in healthy tissue.Finally, we want to understand how genetic heterogeneity relates to the morphological heterogeneity of tumors. This was motivated by a recent clinical study where many discrepancies were observed between the pathological and methylation-based classification of brain tumors. Using TCGA Whole Slide Imaging (WSI) data, we would like to 1) test whether a Convolutional Neural Network is able to accurately predict these methylation classes from the pathology. 2) If yes, identify novel morphological features in the images associated with these classes.
Chair: Dr. Eytan Ruppin Dept. rep: Dr. Soheil Feizi Members: Dr. Sridhar Hannenhalli Dr. Max Leiserson