PLM-ARG

PLM-ARG predicts and classifies antibiotic resistance genes (ARGs) from protein sequences to identify ARGs and assign them to resistance categories for downstream biological analysis.

Key Features:

• Comprehensive Training Data: Trained on over 28,000 ARGs spanning 29 resistance categories.
• High Predictive Accuracy: Achieved Matthew's correlation coefficient (MCC) of 0.983 ± 0.001 in a 5-fold cross-validation.
• Independent Validation: Validated on an independent dataset with MCC = 0.838, outperforming other public ARG prediction tools by 51.8%–107.9%.
• Dual Functionality: Simultaneously detects ARGs and classifies them into resistance categories.
• Practical Applications: Applied to annotate resistance in the UniProt database and to assess the impact of ARGs on environmental microbiota.

Scientific Applications:

• Database annotation: Systematic annotation of ARGs in protein databases such as UniProt.
• Metagenomics and metatranscriptomics: Discovery of previously unrecognized resistance genes in metagenomic and metatranscriptomic datasets.
• Environmental microbiota assessment: Evaluation of ARG prevalence and potential impact on environmental microbial communities.
• Risk assessment: Contribution to risk assessments and management strategies for antibiotic resistance through improved ARG detection.

Methodology:

Uses a pretrained protein language model to learn sequence representations and identify ARGs based on those learned representations rather than direct sequence similarity; performance was evaluated with 5-fold cross-validation and an independent validation dataset.