Understanding Antimicrobial Resistance through Interpretable LLMs

Published: November 2025 | Author: Elias Hossain

Antimicrobial resistance (AMR) has become a defining biomedical challenge of our century, a silent pandemic that grows stronger with every misused antibiotic. While traditional genomic models can classify resistant strains, they often lack interpretability, the ability to explain the reasoning behind predictions. My current research focuses on bridging this gap using Large Language Models (LLMs) fine-tuned for biomedical data, making AMR prediction both accurate and explainable.

      My motivation was simple: prediction alone is not enough. In healthcare,
      every AI decision must be interpretable. A clinician should know why a
      model thinks a strain is resistant.
    

From Observation to Insight

During my early experiments, I noticed that resistance is not just a sequence-level phenomenon. It is ecological, behavioral, and evolutionary. To visualize this, I designed a simplified model of bacterial adaptation that connects exposure, mutation, and spread.

Figure: The progression of antimicrobial resistance, from exposure to global transmission.

The diagram shows how exposure to antibiotics triggers selective pressure. Susceptible bacteria die, while resistant ones survive and replicate. Over generations, these survivors dominate the population, eventually spreading resistance genes across species and even environments. Understanding this progression inspired me to model resistance as an evolving language, one that LLMs could decode through genomic and semantic cues.

LLMs for AMR Reasoning

In my framework, I integrate biomedical LLMs such as BioBERT and PubMedGPT with molecular embeddings to infer resistance pathways. The model does not stop at predicting, it generates human-readable rationales like:

      “Resistance likely arises from beta-lactamase gene mutations impacting
        cefotaxime degradation efficiency.”
    

These insights bridge machine predictions and biological interpretation, helping researchers trace resistance at both the molecular and evolutionary scales.

Why It Matters

Transparent AI models in microbiology can guide clinicians toward better treatment strategies, reduce diagnostic uncertainty, and identify potential drug targets earlier. More importantly, interpretability builds trust, a fundamental requirement for AI integration in healthcare.

From: Elias Hossain
Ph.D. Student, University of Central Florida (UCF)