CONNET: Accurate Genome Consensus in Assembling Nanopore Sequencing Data via Deep Learning

Abstract

Single-molecule sequencing technologies produce much longer reads compared with next-generation sequencing, greatly improving the contiguity of de novo as- sembly of genomes. However, the relatively high error rates in long reads make it challenging to obtain high-quality assemblies. A computationally intensive consensus step is needed to resolve the discrepancies in the reads. Efficient consensus tools have emerged in the recent past, based on partial-order align- ment. In this study, we discovered that the spatial relationship of alignment pileup is crucial to high-quality consensus and developed a deep learning-based consensus tool, CONNET, which outperforms the fastest tools in terms of both accuracy and speed. We tested CONNET using a 903 dataset of E. coli and a 373 human dataset. In addition to achieving high-quality consensus results, CONNET is capable of delivering phased diploid genome consensus. Diploid consensus on the above-mentioned human assembly further reduced 12% of the consensus errors made in the haploid results.