New algorithms for accurate and efficient de-novo genome assembly from long DNA sequencing reads

Abstract

Producing de-novo genome assemblies for complex genomes is possible thanks to long-read DNA sequencing technologies. However, maximizing the quality of assemblies based on long reads is a challenging task that requires the development of specialized data analysis techniques. In this paper, we present new algorithms for assembling long-DNA sequencing reads from haploid and diploid organisms. The assembly algorithm builds an undirected graph with two vertices for each read based on minimizers selected by a hash function derived from the k-mers distribution. Statistics collected during the graph construction are used as features to build layout paths by selecting edges, ranked by a likelihood function that is calculated from the inferred distributions of features on a subset of safe edges. For diploid samples, we integrated a reimplementation of the ReFHap algorithm to perform molecular phasing. The phasing procedure is used to remove edges connecting reads assigned to different haplotypes and to obtain a phased assembly by running the layout algorithm on the filtered graph. We ran the implemented algorithms on PacBio HiFi and Nanopore sequencing data taken from bacteria, yeast, Drosophila, rice, maize, and human samples. Our algorithms showed competitive efficiency and contiguity of assemblies, as well as superior accuracy in some cases, as compared to other currently used software. We expect that this new development will be useful for researchers building genome assemblies for different species.