Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DrugAgent: Multi-Agent Large Language Model-Based Reasoning for Drug-Target Interaction Prediction

Yoshitaka Inoue, Tianci Song, Xinling Wang, Augustin Luna, Tianfan Fu

TL;DR

This work addresses the interpretability and reliability gap in drug-target interaction prediction by introducing DrugAgent, a coordinator-based multi-agent LLM system that fuses ML-based predictions, knowledge-graph reasoning, and literature evidence. The architecture assigns specialized roles to five agents (Coordinator, AI, KG, Search, Reasoning) and uses Chain-of-Thought and ReAct frameworks to deliver transparent, step-by-step predictions. Empirical evaluation on a kinase-inhibitor dataset shows a substantial F1 gain (≈45%) over a non-reasoning baseline, with ablation analyses confirming the critical contributions of the AI, KG, and Search components and highlighting the value of interpretable reasoning for clinical and regulatory contexts. The approach emphasizes interpretability, extensibility, and integration of heterogeneous data sources, offering a practical pathway toward more reliable AI-assisted drug discovery.

Abstract

Advancements in large language models (LLMs) allow them to address diverse questions using human-like interfaces. Still, 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. While drug-target interaction (DTI) prediction is important for drug discovery, existing approaches face challenges due to complex biological systems and the lack of interpretability needed for clinical applications. DrugAgent is a multi-agent LLM system for DTI prediction that combines multiple specialized perspectives with transparent reasoning. Our system adapts and extends existing multi-agent frameworks by (1) applying coordinator-based architecture to the DTI domain, (2) integrating domain-specific data sources, including ML predictions, knowledge graphs, and literature evidence, and (3) incorporating Chain-of-Thought (CoT) and ReAct (Reason+Act) frameworks for transparent DTI reasoning. We conducted comprehensive experiments using a kinase inhibitor dataset, where our multi-agent LLM method outperformed the non-reasoning multi-agent model (GPT-4o mini) by 45% in F1 score (0.514 vs 0.355). Through ablation studies, we demonstrated the contributions of each agent, with the AI agent being the most impactful, followed by the KG agent and search agent. Most importantly, our approach provides detailed, human-interpretable reasoning for each prediction by combining evidence from multiple sources - a critical feature for biomedical applications where understanding the rationale behind predictions is essential for clinical decision-making and regulatory compliance. Code is available at https://anonymous.4open.science/r/DrugAgent-B2EA.

DrugAgent: Multi-Agent Large Language Model-Based Reasoning for Drug-Target Interaction Prediction

TL;DR

This work addresses the interpretability and reliability gap in drug-target interaction prediction by introducing DrugAgent, a coordinator-based multi-agent LLM system that fuses ML-based predictions, knowledge-graph reasoning, and literature evidence. The architecture assigns specialized roles to five agents (Coordinator, AI, KG, Search, Reasoning) and uses Chain-of-Thought and ReAct frameworks to deliver transparent, step-by-step predictions. Empirical evaluation on a kinase-inhibitor dataset shows a substantial F1 gain (≈45%) over a non-reasoning baseline, with ablation analyses confirming the critical contributions of the AI, KG, and Search components and highlighting the value of interpretable reasoning for clinical and regulatory contexts. The approach emphasizes interpretability, extensibility, and integration of heterogeneous data sources, offering a practical pathway toward more reliable AI-assisted drug discovery.

Abstract

Advancements in large language models (LLMs) allow them to address diverse questions using human-like interfaces. Still, 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. While drug-target interaction (DTI) prediction is important for drug discovery, existing approaches face challenges due to complex biological systems and the lack of interpretability needed for clinical applications. DrugAgent is a multi-agent LLM system for DTI prediction that combines multiple specialized perspectives with transparent reasoning. Our system adapts and extends existing multi-agent frameworks by (1) applying coordinator-based architecture to the DTI domain, (2) integrating domain-specific data sources, including ML predictions, knowledge graphs, and literature evidence, and (3) incorporating Chain-of-Thought (CoT) and ReAct (Reason+Act) frameworks for transparent DTI reasoning. We conducted comprehensive experiments using a kinase inhibitor dataset, where our multi-agent LLM method outperformed the non-reasoning multi-agent model (GPT-4o mini) by 45% in F1 score (0.514 vs 0.355). Through ablation studies, we demonstrated the contributions of each agent, with the AI agent being the most impactful, followed by the KG agent and search agent. Most importantly, our approach provides detailed, human-interpretable reasoning for each prediction by combining evidence from multiple sources - a critical feature for biomedical applications where understanding the rationale behind predictions is essential for clinical decision-making and regulatory compliance. Code is available at https://anonymous.4open.science/r/DrugAgent-B2EA.
Paper Structure (22 sections, 7 equations, 1 figure, 1 table)

This paper contains 22 sections, 7 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methods
  4. Overview of DrugAgent
  5. Agent Roles and Responsibilities
  6. Coordinator Agent
  7. AI Agent
  8. KG Agent
  9. Data Sources
  10. Search Agent
  11. Reasoning Agent
  12. Experiment
  13. Performance Comparison of DrugAgent and ablation study
  14. Discussion
  15. Funding
  16. ...and 7 more sections

Figures (1)

  • Figure 1: Multi-agent system architecture for DTI analysis. The system consists of a "Coordinator Agent" that manages three specialized agents for evidence gathering: (1) a Knowledge Graph Agent" accessing biomedical databases (DrugBank, DGIdb, STITCH, CTD) to analyze path-based relationships, (2) an "AI Agent" utilizing the pre-trained DeepPurpose ML model for probability prediction, and (3) a "Search Agent" performing Bing Web Search for literature evidence. The "Reasoning Agent" then integrates this information through CoT and ReAct frameworks to generate final scores with reasoning.