Language-guided domain adaptation improves protein sequence rep-resentations for antimicrobial resistance analysis

BacLLM, a language-guided protein representation framework for WHO priority bacterial pathogens, improves resistance-gene identification and mechanism prediction using annotation priors transferred into sequence encoders.

• Why it matters: AMR complicates infectious disease management, and phenotype-only grouping can miss molecular signals tied to protein function, limiting digital surveillance and therapeutic targeting.
• What they did: BacLLM enhances sequences, enriches UniProt annotations, and uses cross-modal contrastive learning to transfer LLM-encoded annotation priors into a protein sequence encoder.
• The result: It captures resistance-associated mutational signals at single-residue resolution, preserves taxonomic relationships beyond phenotype clinical groupings, recovers TolC-centered functional organization, and improves strain-level phage host matching for candidate prioritization.

AI summary of the abstract · may contain errors · not medical advice · read the paper.