Ontology-Aware Deep Learning Enables Novel Antibiotic Resistance Gene Discovery Towards Comprehensive Profiling of ARGs

Antibiotic resistance genes (ARGs) have emerged in pathogens and spread faster than expected, arousing a worldwide concern. Current methods are suitable mainly for the discovery of close homologous ARGs and have limited utility for discovery of novel ARGs, thus rendering the profiling of ARGs incomprehensive. Here, an ontology-aware deep learning model, ONN4ARG (http://onn4arg.xfcui.com/), is proposed for the discovery of novel ARGs based on multi-level annotations. Experiments based on billions of candidate microbial genes collected from various environments show the superiority of ONN4ARG in comprehensive ARG profiling. Enrichment analyses show that ARGs are both environment-specific and host-specific. For example, resistance genes for rifamycin, which is an important antibacterial agent active against gram-positive bacteria, are enriched in Actinobacteria and in soil environment. Case studies verified ONN4ARG9s ability for novel ARG discovery. For example, a novel streptomycin resistance gene was discovered from oral microbiome samples and validated through wet-lab experiments. ONN4ARG provides a complete picture of the prevalence of ARGs in microbial communities as well as guidance for detection and reduction of the spread of resistance genes.