HyLight: Strain aware assembly of low coverage metagenomes

Abstract

AbstractDifferent strains of identical species can vary substantially in terms of their spectrum of biomedically relevant phenotypes. Reconstructing the genomes of a microbial community at strain level poses major methodical challenges, because relative frequencies of individual strains match the rate of sequencing errors, which hampers the identification of their characteristic genetic variants. While next-generation sequencing (NGS) reads are too short to span complex repetitive regions, the considerably longer third-generation sequencing (TGS) reads are affected by larger sequencing error rates or are just significantly more expensive. Suppressing TGS coverage to limit costs implies losses in terms of the accuracy of the assemblies. Therefore, existing approaches have remained fragmentary: all approaches presented so far agree on losses in strain awareness, accuracy, possibly excessive costs of the assemblies, or combinations thereof.We present HyLight as, to the best of our knowledge, the first metagenome assembly approach that is not affected by any of the aforementioned drawbacks. In experiments, we demonstrate that HyLight assemblies are strain-aware, contiguous, contain little errors, and because operating on low coverage TGS data of the cheap kind, come at drastically reduced costs. HyLight implements hybrid assembly, which exploits the complementarity of TGS and NGS data. For unifying the two types of data, HyLight utilizes strain resolved overlap graphs (OG), which support the accurate reconstruction of the individual members of microbial communities at strain level: HyLight outperforms existing approaches in terms of strain identity preserving sequence by on average 25.53% (across all experiments / data sets: first quartile: 21.53%, median: 26.81%, third quartile: 31.98%), achieving near-complete strain awareness on many data sets. In summary, HyLight appears to implement the first protocol that delivers assemblies that are strain-aware, contiguous and accurate in combination.