This research introduces a novel approach that combines evolutionary algorithms with large language models to dramatically improve molecular discovery efficiency.

• Reduces computational costs by using LLMs to guide chemical space exploration
• Enables more effective optimization of non-differentiable molecular objectives
• Demonstrates superior performance in structure-based drug design tasks
• Achieves up to 4.7x improvement in sample efficiency compared to baseline methods

Why it matters for medicine: This approach could significantly accelerate drug discovery timelines by making the search for candidate molecules more efficient and targeted, potentially reducing development costs and bringing treatments to patients faster.

Efficient Evolutionary Search Over Chemical Space with Large Language Models