Automated Machine Learning Pipeline for Training and Analysis Using Large Language Models

2025-09-25

AI Summary

To address the heavy reliance on human expertise in data generation, model training, and validation for machine-learned interatomic potentials (MLIPs), this work introduces AMLP-Analysis—the first end-to-end automated MLIP pipeline integrating large language model (LLM) agents. Built upon the MACE architecture and the Atomic Simulation Environment (ASE), the LLM agent autonomously configures quantum-chemical calculations, invokes computational codes, and parses outputs, enabling full automation from *ab initio* data generation and model training to molecular dynamics (MD) deployment. Applied to acridine polymorphs, the resulting MLIP achieves mean absolute errors of 1.7 meV/atom (energy) and 7.0 meV/Å (forces), sub-angstrom structural accuracy, and long-term MD stability under NVE and NVT ensembles. This work pioneers deep LLM integration into a closed-loop MLIP development workflow, substantially reducing dependence on domain experts.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-in-the-loop Machine Learning
• Multiagent Systems: Multiagent Learning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Machine learning interatomic potentials (MLIPs) have become powerful tools to extend molecular simulations beyond the limits of quantum methods, offering near-quantum accuracy at much lower computational cost. Yet, developing reliable MLIPs remains difficult because it requires generating high-quality datasets, preprocessing atomic structures, and carefully training and validating models. In this work, we introduce an Automated Machine Learning Pipeline (AMLP) that unifies the entire workflow from dataset creation to model validation. AMLP employs large-language-model agents to assist with electronic-structure code selection, input preparation, and output conversion, while its analysis suite (AMLP-Analysis), based on ASE supports a range of molecular simulations. The pipeline is built on the MACE architecture and validated on acridine polymorphs, where, with a straightforward fine-tuning of a foundation model, mean absolute errors of ~1.7 meV/atom in energies and ~7.0 meV/Å in forces are achieved. The fitted MLIP reproduces DFT geometries with sub-Å accuracy and demonstrates stability during molecular dynamics simulations in the microcanonical and canonical ensembles.

Problem

• Automates MLIP development from dataset creation to validation
• Uses LLM agents for electronic-structure code selection and preparation
• Achieves quantum-accuracy with foundation model fine-tuning on polymorphs

Innovation

• Automated pipeline unifies dataset creation to model validation
• Large-language-model agents assist electronic-structure code operations
• MACE architecture with fine-tuning achieves near-quantum accuracy