MixRx: Predicting Drug Combination Interactions with LLMs

TL;DR

The paper investigates whether large language models can predict multi-drug interaction classes (Additive, Synergistic, Antagonistic) from patient drug histories in emergency medicine, extending prior CancerGPT work beyond oncology. It evaluates GPT-2 and Mistral-7B-Instruct-v0.2 (including fine-tuned variants) on three datasets (SynergxDB, its messy variant, and NIH encounters) using a ground-truth pipeline and a prompt-based framework that encodes pairwise synergy scores. The results show that the Mistral fine-tuned model delivers strong performance (average accuracy ≈ 0.815 with high precision, and 0.82 on standard tests), and maintains robustness to input noise via the Messy dataset, highlighting the potential of LLM-assisted decision support in emergency medicine. The study identifies limitations such as class imbalance and the need for broader validation, and outlines future work including dataset balancing and deeper analysis of model reasoning in real-world settings.

Abstract

MixRx uses Large Language Models (LLMs) to classify drug combination interactions as Additive, Synergistic, or Antagonistic, given a multi-drug patient history. We evaluate the performance of 4 models, GPT-2, Mistral Instruct 2.0, and the fine-tuned counterparts. Our results showed a potential for such an application, with the Mistral Instruct 2.0 Fine-Tuned model providing an average accuracy score on standard and perturbed datasets of 81.5%. This paper aims to further develop an upcoming area of research that evaluates if LLMs can be used for biological prediction tasks.