[Paper Review] LIDDIA: Language-based Intelligent Drug Discovery Agent

Reza Averly, Frazier N. Baker+2 more|ArXiv.org|Feb 19, 2025

Scientific Computing and Data Management3 citations

TL;DR

This paper presents LIDDIA, an autonomous LLM-driven agent for in silico pre-clinical drug discovery that generates and optimizes novel drug candidates across multiple targets, achieving high-quality, diverse outputs with strong target success.

ABSTRACT

Drug discovery is a long, expensive, and complex process, relying heavily on human medicinal chemists, who can spend years searching the vast space of potential therapies. Recent advances in artificial intelligence for chemistry have sought to expedite individual drug discovery tasks; however, there remains a critical need for an intelligent agent that can navigate the drug discovery process. Towards this end, we introduce LIDDIA, an autonomous agent capable of intelligently navigating the drug discovery process in silico. By leveraging the reasoning capabilities of large language models, LIDDIA serves as a low-cost and highly-adaptable tool for autonomous drug discovery. We comprehensively examine LIDDIA , demonstrating that (1) it can generate molecules meeting key pharmaceutical criteria on over 70% of 30 clinically relevant targets, (2) it intelligently balances exploration and exploitation in the chemical space, and (3) it identifies one promising novel candidate on AR/NR3C4, a critical target for both prostate and breast cancers. Code and dataset are available at https://github.com/ninglab/LIDDiA

Motivation & Objective

• Motivate the need for autonomous, intelligent routing of the drug discovery workflow.
• Propose a modular agent architecture that combines reasoning, execution, evaluation, and memory with grounding tools.
• Demonstrate that the agent generates high-quality, novel molecules across multiple targets and properties.
• Show that the agent balances exploration and exploitation and identifies promising candidates for key targets like EGFR.

Proposed method

• Introduce a four-component architecture: Reasoner, Executor, Evaluator, and Memory.
• Ground Reasoner and Evaluator in Claude 3.5/related LLMs with task-specific prompts.
• Use Pocket2Mol for generation, GraphGA for optimization, and a screening processor for selection.
• Evaluate molecules on QED, Lipinski’s Rule of Five (LRF), SAS, Vina affinity (VNA), and novelty (NVT).
• Measure Diversity (DVS) and Target Success Rate (TSR) across 30 protein targets with known structures and drugs.

Experimental results

Research questions

• RQ1 Can an autonomous LLM-based agent generate diverse, high-quality drug candidates across multiple targets?
• RQ2 How well does the agent balance exploration and exploitation in chemical space to improve hit identification and lead optimization?
• RQ3 Is the agent capable of identifying promising candidates for clinically relevant targets (e.g., EGFR) with favorable property profiles?

Key findings

• LIDDIA achieves a TSR of 73.3% across 30 targets, outperforming baseline methods.
• On average, 85% of generated molecules per target are high quality (HQ) across key properties.
• Molecules generated by LIDDIA show high novelty and strong VNA (binding affinity) profiles, outperforming baselines in VNA.
• LIDDIA combines de novo generation and iterative optimization to deliver molecules with HQ across QED, LRF, SAS, and VNA.
• Case study on EGFR shows promising candidates with profiles comparable to approved drugs.