Statistical methods for genomic analysis of heterogeneous tumors

Project Summary In most cancers, heterogeneous cell composition within and between tumor samples is mirrored in complex variations at a molecular level. This molecular complexity includes both transcriptional variation and genomic complexity, since tumors continually evolve and acquire new mutations. Therefore, to further our understanding

of tumor evolution, it is essential to study the evolutionary dynamics between cancer genomes and transcriptomes. However, due to the complex interplay between cancer cells and their environment, these dynamics are still poorly understood, which presents a major bottleneck for the advancement of clinical

management and treatment of cancer patients. Recent multi-region matched DNA/RNA sequencing studies have made significant advances in our understanding of cancer evolutionary dynamics. However, the analytical tools used in these studies were limited to one molecular data type at a time, representing a missed opportunity for

novel biological discovery. The overall objective of this proposal is to 1) quantify, at scale, the evolutionary dynamics between genomic and transcriptomic variations in cancer cells; and 2) link this quantity to cancer prognosis and therapeutic response. On the methodological side, we will develop a suite of integrative

deconvolution models for matched genomic and transcriptomic data types. Multiple angles to approach the matched data will be evaluated in separate statistical models. On the applied side, we will focus on the clinical impact of such models on the treatment of prostate (PCa) and thyroid cancers (THCa). These two cancers rank

3rd and 12th in prevalence and are projected by the CDC to present a total of 292,810 new cases in 2021 in the US. For both cancers, overtreatment is the most clinically urgent problem since there is no clear method to differentiate low-risk patients from those at high risk. We hypothesize that biomarkers informed by tumor

evolutionary trajectory may identify patients who do not need further treatment. Identification of these biomarkers would significantly improve the efficiency of clinical practice. Our research group consists of experienced investigators with complementary expertise in tumor heterogeneity and clinical management of cancers.

Together, we propose the following Aims: 1. Develop integrative deconvolution models to study the evolution of transcriptomes in cancer cells, 1A) at the cell-type and gene levels, 1B) at single-nucleotide-variant level, 1C) genomic heterogeneity over a multi-sample design, and 1D) transcriptomic heterogeneity over a multi-sample

design; 2. Apply integrative models to cancer patients for biomarker discovery in 2A) high-risk prostate cancer and 2B) high-risk thyroid cancer; 3. Develop user-friendly and computationally efficient software tools for cancer genomics. The proposed methods and tools are expected to open new avenues to discovery by enabling

comprehensive profiling of tumor cell types over evolution and associating these values with clinical outcomes. Our proof-of-principle investigation of prostate and thyroid cancers has the potential to identify new integrative biomarkers predictive of cancer prognosis and response to treatment.