Scalable methods for storage, processing and analysis of sequencing datasets

Abstract

Massive amounts of next-generation sequencing (NGS) reads generated from sequencing machines around the world have revolutionized biotechnology enabling wide-scale disease and variation studies, personalized medicine and helping us understand our evolutionary history. However, the amount of sequencing data generated every day increases at an exponential rate posing an imminent need for smart algorithmic solutions to handle massive sequencing datasets and efficiently extract the useful knowledge within them. This thesis consists of four research contributions on these two fronts. First, we present a computational framework that leverages the redundancy within large genomic datasets for performing faster read-mapping while improving sensitivity. Second, we describe a lossy compression method for quality scores within sequencing datasets that strikingly improves the downstream accuracy for genotyping. Third, we introduce a Bayesian framework for accurate diploid and polyploid haplotype reconstruction of an individual genome using NGS datasets. Lastly, we extend this haplotype reconstruction framework to high-throughput transcriptome sequencing datasets.