Better Identification of Repeats in Metagenomic Scaffolding

Genomic repeats are the most important challenge in genomic assembly. While for single genomes the effect of repeats is largely addressed by modern long-read sequencing technologies, in metagenomic data intra-genome and, more importantly, inter-genome repeats continue to be a significant impediment to effective genome reconstruction. Detecting repeats in metagenomic samples is complicated by characteristic features of these data, primarily uneven depths of coverage and the presence of genomic polymorphisms. The scaffolder Bambus 2 introduced a new strategy for repeat detection based on the betweenness centrality measure - a concept originally used in social network analysis. The exact computation of the betweenness centrality measure is, however, computationally intensive and impractical in large metagenomic datasets. Here we explore the effectiveness of approximate algorithms for network centrality to accurately detect genomic repeats within metagenomic samples. We show that an approximate measure of centrality achieves much higher computational efficiencies with a minimal loss in the accuracy of detecting repeats in metagenomic data. We also show that the combination of multiple features of the scaffold graph provides a more effective strategy for identifying metagenomic repeats, significantly outperforming all other commonly used approaches.