Development and Application of Methods to Discover Cancer-Associated Transcript Variants.

Abstract

Cancer is and has long been a major threat to human health, and in seeking to better treat cancer, we seek first to better understand cancer. Consequently, the current era of cancer research has aimed to catalog the full range of molecular abnormalities in cancer's genome, epigenome, transcriptome, and proteome. Next-generation short read sequencing has empowered these cataloging efforts, but requires sophisticated algorithms to fully harness, particularly in the case of splicing and transcript variation. The aim of this dissertation was to address this need by establishing and applying novel methods to analyze RNA sequencing data in cancer.

In chapter one, we present Oculus, a software package that attaches to standard aligners and exploits read redundancy by performing streaming compression, alignment, and decompression of input sequences. This nearly lossless process (> 99.9%) led to alignment speedups of up to 270% across a variety of data sets.

In chapter two, we profile performance characteristics of two-pass alignment, which separates splice junction discovery from quantification. Across a variety of transcriptome sequencing datasets, two-pass alignment improved quantification of at least 94% of simulated novel splice junctions, and provided as much as 1.7-fold deeper median read depth over those splice junctions. Two-pass alignment promises to advance quantification and discovery of novel splicing events.

In chapter three, we present a novel bioinformatics pipeline to analyze splicing and transcript variation from cancer transcriptome data, using splice junction read depth, and correlative analysis to circumvent known biases such as tumor content. We demonstrate the value of this pipeline through application to the oncogenes MET and ALK.

Finally, in chapter four, we present the application of our transcript variant calling pipeline to transcriptome data from prostate cancer. We present several recurrently differentially spliced genes which are not attributable to noise or bias and may serve as novel biomarkers, evidence for transcript variants of the androgen receptor, and an apparent genome-wide pattern of alternative transcription start site usage.