HMD-ARG: hierarchical multi-task deep learning for annotating antibiotic resistance genes.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Antibiotic resistance is a critical global health issue, causing 700,000 deaths annually.
• Antibiotic resistance genes (ARGs) spread rapidly between environmental and human reservoirs, reducing antibiotic effectiveness.
• Current ARG identification methods, primarily sequence alignment-based, struggle with novel ARGs and incomplete databases.

Purpose of the Study:

• To develop an advanced computational framework for accurate and comprehensive identification of antibiotic resistance genes (ARGs).
• To overcome limitations of existing methods by enabling identification of novel ARGs and detailed annotation of their properties.

Main Methods:

• Proposed a Hierarchical Multi-task Deep learning framework for ARG annotation (HMD-ARG).
• HMD-ARG processes raw sequence data to simultaneously predict ARG presence, antibiotic class, resistance mechanism, and gene mobility (intrinsic vs. acquired).
• For beta-lactamase predictions, HMD-ARG further identifies specific subclasses.

Main Results:

• HMD-ARG demonstrated superior performance compared to state-of-the-art methods in identifying ARGs and their characteristics.
• Validation included cross-fold, third-party datasets (human gut microbiota), wet-experimental functional assays, and structural investigations.
• The method proved effective and robust in identifying multiple ARG properties from raw sequence encoding without database queries.

Conclusions:

• HMD-ARG provides detailed annotations of ARGs, including resistance class, mechanism, and mobility.
• This deep learning framework is a powerful tool for identifying ARGs and mitigating the global threat of antibiotic resistance.
• The HMD-ARG method and database are publicly available for research use.