New Framework Combines Large Language Models with Answer Set Programming for Explainable Disease Diagnosis

Global: New Framework Combines Large Language Models with Answer Set Programming for Explainable Disease Diagnosis

Researchers Ioanna Gemou and Evangelos Lamprou presented a proof‑of‑concept system, named McCoy, that merges large language models (LLMs) with answer set programming (ASP) to generate interpretable disease diagnoses. The work was submitted to arXiv on 30 December 2025 and targets the challenge of delivering accurate, timely predictions while preserving explanatory transparency.

Background

Accurate disease prediction is recognized as essential for early intervention and effective treatment. Prior applications of symbolic artificial intelligence in healthcare have been constrained by the labor‑intensive process of building high‑quality knowledge bases, limiting broader adoption.

Methodology

McCoy orchestrates an LLM to convert relevant medical literature into ASP rules. These rules are then combined with individual patient data and processed by an ASP solver, which derives a diagnosis through logical inference. The integration leverages the pattern‑recognition strength of LLMs and the rigorous, traceable reasoning offered by ASP.

Interpretability

Because ASP produces explicit logical models, the resulting diagnoses can be examined step‑by‑step, providing clinicians with a clear rationale for each recommendation. This contrasts with the often opaque decision pathways of purely statistical or neural approaches.

Preliminary Findings

Initial experiments on small‑scale disease diagnosis tasks demonstrated that McCoy achieved strong performance metrics, matching or exceeding baseline models while maintaining full explainability. The authors note that these results are early and derived from limited datasets.

Implications and Future Directions

If the approach scales, it could lower the barrier for deploying symbolic AI techniques in clinical settings, offering a blend of accuracy and transparency. Ongoing work will focus on expanding the knowledge base, testing on larger patient cohorts, and evaluating real‑world clinical utility.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.