Graph-Based Approaches Significantly Improve the Recovery of Antibiotic Resistance Genes From Complex Metagenomic Datasets.

The lack of control over the usage of antibiotics leads to propagation of the microbial strains that are resistant to many antimicrobial substances. This situation is an emerging threat to public health and therefore the development of approaches to infer the presence of resistant strains is a topic of high importance. The resistome construction of an isolate microbial species could be considered a solved task with many state-of-the-art tools available. However, when it comes to the analysis of the resistome of a microbial community (metagenome), then there exist many challenges that influence the accuracy and precision of the predictions. For example, the prediction sensitivity of the existing tools suffer from the fragmented metagenomic assemblies due to interspecies repeats: usually it is impossible to recover conservative parts of antibiotic resistance genes that belong to different species that occur due to e.g., horizontal gene transfer or residing on a plasmid. The recent advances in development of new graph-based methods open a way to recover gene sequences of interest directly from the assembly graph without relying on cumbersome and incomplete metagenomic assembly. We present GraphAMR-a novel computational pipeline for recovery and identification of antibiotic resistance genes from fragmented metagenomic assemblies. The pipeline involves the alignment of profile hidden Markov models of target genes directly to the assembly graph of a metagenome with further dereplication and annotation of the results using state-of-the art tools. We show significant improvement of the quality of the results obtained (both in terms of accuracy and completeness) as compared to the analysis of an output of ordinary metagenomic assembly as well as different read mapping approaches. The pipeline is freely available from https://github.com/ablab/graphamr.