RAG-Enhanced Collaborative LLM Agents for Drug Discovery

TL;DR

This work tackles the challenge of applying large language models to drug discovery without costly domain-specific fine-tuning, addressing heterogeneous biochemical data and open-ended questions. It introduces CLADD, a retrieval-augmented, multi-agent framework with Planning, Knowledge Graph, and Molecular Understanding teams that dynamically integrates external annotations and a knowledge graph via anchoring and 2-hop relations, all without LLM fine-tuning. Across drug-target prediction, property-specific captioning, and biological activity prediction, CLADD achieves state-of-the-art zero-shot performance compared to general-purpose and domain-tuned LLMs, while providing interpretable agent interactions. The framework enables rapid incorporation of new data and diverse evidence, with public code to support reproducibility and broader adoption in AI-assisted drug discovery.

Abstract

Recent advances in large language models (LLMs) have shown great potential to accelerate drug discovery. However, the specialized nature of biochemical data often necessitates costly domain-specific fine-tuning, posing major challenges. First, it hinders the application of more flexible general-purpose LLMs for cutting-edge drug discovery tasks. More importantly, it limits the rapid integration of the vast amounts of scientific data continuously generated through experiments and research. Compounding these challenges is the fact that real-world scientific questions are typically complex and open-ended, requiring reasoning beyond pattern matching or static knowledge retrieval.To address these challenges, we propose CLADD, a retrieval-augmented generation (RAG)-empowered agentic system tailored to drug discovery tasks. Through the collaboration of multiple LLM agents, CLADD dynamically retrieves information from biomedical knowledge bases, contextualizes query molecules, and integrates relevant evidence to generate responses - all without the need for domain-specific fine-tuning. Crucially, we tackle key obstacles in applying RAG workflows to biochemical data, including data heterogeneity, ambiguity, and multi-source integration. We demonstrate the flexibility and effectiveness of this framework across a variety of drug discovery tasks, showing that it outperforms general-purpose and domain-specific LLMs as well as traditional deep learning approaches. Our code is publicly available at https://github.com/Genentech/CLADD.

Figures (13)

• Figure 1: Overview of CLADD.
• Figure 2: Ablation studies.(a) On model components. (b) On external knowledge.
• Figure 3: Example of collaboration between agents in CLADD (on the drug-target prediction task). Red represents adrenergic receptors, yellow represents histamine receptors, and green represents dopamine receptors (The full version in Appendix \ref{['app: additional experiments']}).
• Figure 4: (a) After CLADD (or baseline models) generates a property-specific caption, (b) SciBERT is used for evaluation. In other words, fine-tuning SciBERT is not part of CLADD; it is only used for evaluation purposes.
• Figure 5: Data analysis on PubChem database.