HMD-ARG

HMD-ARG employs a hierarchical multi-task deep learning framework to annotate antibiotic resistance genes (ARGs) from raw protein sequences and predict resistance class, resistance mechanism, and gene mobility.

Key Features:

• Multi-task Annotation: Determines whether a given protein sequence is an ARG and annotates its resistance to specific antibiotic families.
• Detailed Mechanism Identification: Distinguishes resistance mechanisms and classifies genes as intrinsic or acquired.
• Beta-lactamase Subclass Prediction: Predicts the subclass of beta-lactamase for ARGs associated with beta-lactam antibiotics.
• Hierarchical Multi-task Approach: Integrates hierarchical multi-task predictions to provide simultaneous outputs for resistant antibiotic class, resistant mechanism, and gene mobility.
• End-to-end Raw Sequence Encoding: Processes raw sequence encodings without relying on queries to existing sequence databases.

Scientific Applications:

• Benchmarking and Validation: Demonstrated superior performance versus state-of-the-art methods using cross-fold validation and third-party dataset assessments in human gut microbiota.
• Experimental Validation Support: Predictions have been subjected to wet-experimental functional validations.
• Structural Investigation: Facilitates identification of conserved sites for downstream structural investigations of predicted ARGs.

Methodology:

Hierarchical multi-task end-to-end deep learning model that processes raw sequence encodings and simultaneously predicts multiple ARG properties without querying existing sequence databases.