New approaches to big data in genome sequencing (360G-Wellcome-109082_Z_15_A)

The ever increasing throughput and reducing cost of DNA sequencing continue to challenge computational data analysis associated with genome sequencing, requiring novel algorithmic and software solutions that take advantage of structure in the data. We will develop new efficient computational solutions to handle and analyse very large genome sequence data sets. Specifically, we plan to develop software to address the following topics, in the context of the human Haplotype Reference Consortium with tens of thousands of whole human genome sequences, and also large non-human vertebrate reference genome sequencing projects. We will develop compressed data structures to store deep population resequencing data, while supporting fast genetic analyses that work directly from the compressed data. We will extend these methods to new reference structures based on variation graphs. We will develop new approaches to long range genome sequence assembly, using long reads to scaffold deep short read assemblies, and using linkage disequilibrium structure in population data to scaffold assemblies, incorporating methods from topic (1). We will use machine learning approaches to address statistical problems such as indel and structural variant calling that are hard to model exactly, using large simulated and real data sets and methods from (1) and (2).