Identifying prognostic biomarkers for cancer using gene expression data

Abstract

With the advancements in next-generation sequencing technologies, there is an unprecedented amount of tumor genomic and transcriptomic data available for research. This data availability opens the door for using data-driven techniques to identify biomarkers that can provide useful information about a patient's tumor, such as progression risk and potential sensitivity to treatment. Given the heterogeneity of the disease, the availability of such patient-specific information to clinicians will enable personalized care and improve clinical outcomes. In this thesis we utilize gene expression data from The Cancer Genome Atlas to identify prognostic biomarkers and predictors of potential response to immunotherapy in lung adenocarcinoma (LUAD) and metastatic skin cutaneous melanoma (SKCM). We utilize various statistical and machine learning techniques in the identification of such biomarker genes, and we analyze the final candidate genes for their utility in characterizing the immune response to the tumor and in stratifying patients by survival time. For both LUAD and SKCM we identified genes which can be reliable predictors of patient survival and whose enrichment correlates with immune cell abundance in the tumor micro-environment, specifically with CD4 and CD8 T-cells, and macrophages. We also found that some of the genes we identified represent known cancer hallmarks. Finally, a subset of the genes we identified have been shown to be predictive of response to immunotherapy in melanoma patients in a previous study. These results show that the identified genes may have biological and clinical significance in the context of cancer. Furthermore, the methodology we used to identify those biomarkers was not dependent on any disease-specific information, and can be adjusted for utility in other cancer types or even other diseases.