An efficient overlap graph coarsening approach for modeling short reads

Abstract

Next generation sequencing has quickly emerged as the most exciting yet challenging computational problem in Bioinformatics. Current sequencing technologies are capable of producing several hundreds of thousands to several millions of short sequence reads in a single run. However, current methods for managing, storing, and processing the produced reads remain for the most part simple and lack the complexity needed to model the produced reads efficiently and assemble them correctly. These reads are produced at a high coverage of the original target sequence such that many reads overlap. The overlap relationships are used to align and merge reads into contiguous sequences called contigs. In this paper, we present an overlap graph coarsening scheme for modeling reads and their overlap relationships. Our approach is different from previous read analysis and assembly methods that use a single graph to model read overlap relationships. Instead, we use a series of graphs with different granularities of information to represent the complex read overlap relationships. We present a new graph coarsening algorithm for clustering a simulated metagenomics dataset at various levels of granularity. We also use the proposed graph coarsening scheme along with graph traversal algorithms to find a labeling of the overlap graph that allows for the efficient organization of nodes within the graph data structure.