UC Santa Cruz

Predicting the most effective cancer therapy for patients is a challenging yet very important task. In my doctoral thesis, I describe new insights gained from using transcriptional signatures from gene expression data of tumors and the tumor microenvironment. Out of different multi-omics data types, gene expression is found to be the most useful in predicting cancer drug sensitivity in a data set of cancer cell lines. Gene expression data can also be used to predict the presence of cancer driver events, genetic abnormalities responsible for tumor growth and progression. I describe the detection of rare genomic driver events found by association with known driver events using transcriptional signatures.

Cancers are traditionally classified into types and subtypes by the organ- and cell-of-origin. However, more and more cancer subtypes are now being defined on a molecular basis using for example gene expression or mutation data. I perform a meta-analysis of molecular subtype classifiers for 26 different cancer cohorts that demonstrates which aspects of the input samples and input data are important to build an accurate molecular subtype classifier. In advanced prostate cancer, I use transcriptional signatures to reliably classify samples into subtypes. The gene expression data, in combination with histological review, is able to define the most at-risk patients in this cohort. However, I show that the classification of cancers into distinct subtypes is not applicable to all samples in this cohort, because they exist on a continuous spectrum between the subtypes - a finding that I was able to recapitulate in a study of lung cancer samples. I define and describe this continuum between subtypes of advanced prostate cancer using gene expression data.

The tumor microenvironment, the normal cells that mix with cancer cells to form a tumor, plays an important role in cancer progression and treatment response. I built a landscape of about 10,000 cancer samples from immune cell infiltration estimated by deconvolution of gene expression profiles. However, immune cells are not the only cell types in the tumor microenvironment. I present a comprehensive deconvolution analysis of tumor patient samples using a cell type library defined from single-cell RNA sequencing data. These cell type estimates enable the detection of a pan-cancer high-risk sample group that is not detected by traditional gene expression analysis.