LIDDIA: Language-based Intelligent Drug Discovery Agent.

Area of Science:

• Artificial intelligence in chemistry
• Computational drug discovery
• Medicinal chemistry

Background:

• Drug discovery is a lengthy, costly, and intricate process heavily reliant on human expertise.
• Existing artificial intelligence (AI) tools accelerate specific tasks but lack end-to-end navigation capabilities.
• A need exists for an intelligent system to autonomously guide the drug discovery pipeline.

Purpose of the Study:

• To introduce LIDDiA, an autonomous AI agent designed for in silico drug discovery.
• To demonstrate LIDDiA's capability in navigating the entire drug discovery process.
• To present LIDDiA as a low-cost, adaptable solution leveraging large language models.

Main Methods:

• Development of LIDDiA, an autonomous agent utilizing large language models for reasoning.
• In silico evaluation of LIDDiA across 30 clinically relevant targets.
• Assessment of LIDDiA's ability to balance exploration and exploitation in chemical space.

Main Results:

• LIDDiA successfully generated molecules meeting pharmaceutical criteria for over 70% of tested targets.
• The agent demonstrated intelligent exploration and exploitation within the chemical search space.
• A novel drug candidate targeting AR/NR3C4, crucial for prostate and breast cancers, was identified.

Conclusions:

• LIDDiA represents a significant advancement in autonomous drug discovery.
• The AI agent offers a cost-effective and adaptable approach to identifying potential therapeutics.
• LIDDiA successfully identified a promising candidate for AR/NR3C4, highlighting its potential in oncology drug development.