Large language models for molecular design: bridging the gap between chemical syntax and biological semantics

Area of Science:

• Artificial Intelligence in Drug Discovery
• Computational Chemistry
• Bioinformatics

Background:

• Large language models (LLMs) are increasingly utilized in early drug discovery.
• Traditional methods often struggle with the complexity of chemical syntax and biological semantics.
• Bridging the gap between chemical representations and biological function is a key challenge.

Purpose of the Study:

• To review the evolution of LLMs in drug discovery, from sequence-based models to autonomous systems.
• To analyze various LLM paradigms for hit identification and lead optimization.
• To discuss advancements in structure-aware and multimodal frameworks for molecular design.

Main Methods:

• Review of existing literature on LLMs in drug discovery.
• Analysis of diverse LLM paradigms including de novo design and multi-objective optimization.
• Exploration of structure-aware, multimodal, and autonomous agent frameworks.

Main Results:

• LLMs are evolving from sequence-based models to sophisticated autonomous discovery systems.
• Structure-aware and multimodal frameworks enhance molecular design by incorporating 3D information.
• Autonomous agents can orchestrate closed-loop design-make-test-analyze (DMTA) cycles.

Conclusions:

• There is a significant gap between computational predictions and experimental validation in LLM-driven drug discovery.
• Future directions include hybrid neuro-symbolic architectures and unified foundation models.
• Robust and autonomous molecular design requires further advancements in LLM integration and validation.