Chat-Driven Computational (Bio)chemistry: Using LLM Agents to Accelerate Bio- and Chemoinformatics

Area of Science:

• Computational chemistry
• Bioinformatics
• Software engineering

Background:

• Large-language models (LLMs) are increasingly used in software engineering.
• LLMs can autonomously manage codebases, improving complex software.
• They offer potential for developing and enhancing bio- and chemoinformatic tools.

Purpose of the Study:

• To demonstrate the use of LLMs for modernizing and optimizing computational chemistry tools.
• To address specific performance limitations in the AMBER molecular dynamics suite's LEaP program.

Main Methods:

• Utilized a large-language model to identify and rectify issues in the AMBER LEaP program.
• Implemented an optimized unit merge algorithm and 64-bit indexing to overcome limitations.
• Tested the LLM's performance on parametrization tasks for molecular dynamics simulations.

Main Results:

• The LLM resolved an O(N^2) merge routine and a 32-bit integer overflow.
• Parametrization time was reduced by over 10-fold for mid-sized systems.
• The optimized software now supports parametrization of multimillion-molecule systems.

Conclusions:

• LLM agents can effectively modernize and optimize computational (bio)chemistry tools.
• This advancement allows for the simulation of significantly larger molecular systems.
• New challenges arise in software provenance and the evolving roles of developers.