In silico dissection of transcriptomes with a tumor immunology focus

This PhD thesis studies dimensionality reduction using independent component analysis as a methodology to conduct meta-analysis of publicly available genome-wide mRNA expression studies. Expression profiles of multiple studies were downloaded, quality controlled, and pre-processed together to obtain normalized gene level mRNA expression measurements. Dimensionality reduction was applied to obtain a reduced representation of the expression profiles in the form of unit vectors that capture latent expression patterns. This thesis studied datasets derived from human and mouse Affymetrix mRNA microarrays (GPL570, GPL1261) and from human Illumina RNA sequencing technologies. Publicly available web repositories utilized included the Gene Expression Omnibus (GEO), The Cancer Genome Atlas (TCGA), The Cancer Cell line Encyclopedia (CCLE), Genomics Drug Sensitivity in Cancer (GDSC), and the European Nucleotide Archive (ENA). These datasets contained gene or single transcript level measurements ranging from 18,425 to 58,433 measurements. The biggest dataset contained 106,462 samples. Latent expression patterns were used to construct gene to gene co-expression models for gene function prediction applications. Additionally, selection strategies were developed to isolate latent expression patterns that capture the downstream mRNA expression effect of somatic copy number alteration, metabolic or immune biological processes. Downstream analyses focused on studying the expression patterns present in tumor samples of patients with cancer. These analyses aimed at understanding of the anti-tumor specific immune response triggered with immune checkpoint inhibitor treatment with monoclonal antibodies. A chapter of this thesis also describes the development of a visualization for the European Society for Medical Oncology Magnitude of Clinical Benefit Scale (ESMO-MCBS)