With the advent of next generation sequencing technologies, it has become increasingly evident that tumors, previously thought to be similar on a histological level, exhibit a wide range of inter and intra-tumor genetic heterogeneity. Hence, there is an unmet need for computational methods to characterize this heterogeneity from different perspectives using high dimensional cancer genomics datasets. In this PhD work, we developed a suite of integrative computational techniques to characterize inter and intra-tumor genetic heterogeneity from a variety of high dimensional genomics datasets in order to obtain new biological insights and predict patient clinical outcomes.

First, we developed new computational frameworks for the functional annotation of genes and transcriptional signaling networks with direction of signal flow and signs given genome-wide transcriptomics and genetic perturbation data. Such annotations are key to building an accurate computational model of tumor cells.

Second, we analyzed genome-wide mutation, copy number and transcriptome data of melanoma patients from The Cancer Genome Atlas (TCGA) to study the effects of increasing intra-tumor genetic heterogeneity on the host immune response. This study was motivated by observations that melanoma patients with similar mutation burden levels have varied responses to immune checkpoint blockade therapy. Overall, the results of our computational analysis were experimentally validated in a mouse model, thereby providing additional mechanistic insights into how intra-tumor genetic heterogeneity impacts patient responses to immune checkpoint blockade therapy.

Third, analyzing genome-wide copy number, transcriptome and methylation data of cancer and normal tissues using statistical and machine learning techniques we find that hardwiring of normal chromosome-wide gene expression levels is an additional factor driving the selection of cancer type-specific chromosomal aneuploidies.

Lastly, we developed a new computational tool: CODEFACS (COnfident Deconvolution For All Cell Subsets) and a supporting statistical framework LIRICS (LIgand Receptor Interactions between Cell Subsets) that enables an averaged “virtual single-cell” characterization of the tumor microenvironment (TME) of each sample from bulk transcriptomic data and discovery of clinically relevant cellular crosstalk. Using 15 benchmark test datasets, we first demonstrate that CODEFACS substantially improves our ability to reconstruct cell-type-specific transcriptomes from individual bulk samples, compared to the state-of-the-art method, CIBERSORTx. Next, analyzing the TCGA using CODEFACS + LIRICS, we uncovered a shared repertoire of cell-cell interactions that specifically occur in TME of mismatch-repair-deficient tumors and explain their high response rates to immune checkpoint blockade treatment. These results point to specific T-cell co-stimulating interactions in the TME that enhance immunotherapy responses independent of tumor mutation burden levels. Finally, using machine learning techniques, we identified a subset of cell-cell interactions that predict patient responses to immune checkpoint blockade therapy in melanoma better than recently published bulk transcriptomics-based signatures.

 

                           Chair:              Dr. Soheil Feizi                                  
                          Co-Chair:        Dr. Eytan Ruppin 
                          Dean's rep:      Dr. Najib El-Sayed
                          Members:         Dr.  Max Leiserson