DrugAgent: Multi-Agent Large Language Model-Based Reasoning for Drug-Target Interaction Prediction and Repurposing
Keywords: Multi-agent, Drug-target interaction, drug-protein binding prediction, Large Language Models
Abstract: Advancements in large language models (LLMs) allow them to address a wide set of questions from diverse topics using human-like text interfaces, but limitations in their training prevent them from answering accurately in scenarios that could benefit from multiple perspectives. Multi-agent systems allow the resolution of questions to enhance result consistency and reliability. Here we create a multi-perspective (i.e., unstructured text, structured knowledge graph, and Machine Learning (ML) prediction) multi-agent LLM system. We apply this system to the biologically inspired problem of predicting drug-target interaction. Our system uses a coordinator agent to assign and integrate results for tasks given to three specialized agents: an AI agent for ML predictions, a knowledge graph (KG) agent for KG retrieval, and a search agent for web-based information retrieval.
We conducted experiments using our LLM-based system for predicting drug-target interaction constants that reflect binding affinities using the BindingDB dataset. Our multi-agent LLM method significantly outperformed GPT-4 across multiple evaluation metrics by a significant margin. An ablation study revealed the contributions by each agent; ranked in terms of a contribution: the AI agent (i.e., ML prediction) was the most important followed by the KG agent then the search agent. The large contribution by the AI agent highlights the importance of LLM tool use in addressing questions that may not be part of text corpora. While our use case was related to biology, our presented architecture is applicable to other integrative prediction tasks. Code is available https://anonymous.4open.science/r/DrugAgent-2BB7/
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12495
Loading