Abstract

Clinical and biological data is growing drastically in size and diversity. This development has driven impressive advances in algorithms. It also has made it clear that big data alone does not necessarily allow to forego advanced statistical methods, such as Bayesian Hidden Markov Models (HMM).

Using compression to accelerate computations has been a standard approach for biological sequences. Here we introduce a full Bayesian approach capable of identifying Copy Number Variants (CNV) obtained from whole-genome sequencing experiments even on laptop computers. The novel combination of Gibbs-sampling for HMM with continuous observations and dynamic Wavelet compression accelerates forward-backward Gibbs (FBG) to the point of matching and exceeding the speed of maximum-likelihood methods. At the same time it improves the convergence of the sampler. Current work focuses on extending the method to joint analysis of CNV data in trios, pedigrees, or across multiple conditions, which can provide posterior probabilities, for example, for de novo CNVs in children affected with a specific disease.

For the pre-processing of High-Throughput Sequencing (HTS) data, we introduce our compressive genomics approach, which finds reduced and compressed representations of sequencing data on which downstream algorithms—such as read mappers, error correction tools and variant finders—can operate directly and efficiently.

Prof. Alexander Schliep – Biography

Alexander Schliep earned his PhD degree in computer science at the Universität zu Köln, jointly supervised by David C. Torney at Los Alamos National Laboratory. From 2002 he was the group leader of the Bioinformatics Algorithms Group at the Max Planck Institute for Molecular Genetics in Berlin and moved in 2009 as an associate professor for Computer Science and Quantitative Biology to Rutgers University, New Brunswick, NJ. In 2016 he moved to the department of Computer Science and Engineering (CSE) which is a joined department of Chalmers and Gothenburg University. His main research interest are statistical bioinformatics methods at genome-scale.