Agentic Design of Compositional Descriptors via Autoresearch for Materials Science Applications

TL;DR

This paper demonstrates an autoresearch framework where an AI agent generates and evaluates composition-based descriptors for materials property prediction, outperforming traditional methods and producing interpretable features.

Contribution

The introduction of Automat, an autoresearch system using large language models to design chemically interpretable descriptors without manual feature engineering.

Findings

Automat improves prediction accuracy over baseline methods.

Generated descriptors are chemically interpretable.

Current limitations include descriptor redundancy and sensitivity to greedy expansion.

Abstract

Autoresearch offers a flexible paradigm for automating scientific tasks, in which an AI agent proposes, implements, evaluates, and refines candidate solutions against a quantitative objective. Here, we use composition-based materials-property prediction to test whether such agents can perform a task beyond model selection and hyperparameter optimization: the design of input descriptors. We introduce Automat, an autoresearch framework where a coding agent based on a large language model generates composition-only descriptors for chemical compounds and evaluates them using a random forest workflow. The agent is restricted to information derivable from chemical formulas and iteratively proposes, implements, and tests chemically motivated descriptor strategies. We apply Automat, with OpenAI Codex using GPT-5.5 as the coding agent, to the prediction of experimental band gaps in inorganic materials and Curie temperatures in ferromagnetic compounds. In both tasks, Automat improves over fractional-composition, Magpie, and combined fractional-composition/Magpie baselines, while producing descriptor families that are chemically interpretable. These results provide a demonstration that autoresearch agents can generate competitive, task-specific materials descriptors without manual feature engineering during the run. They also reveal current limitations, including descriptor redundancy, sensitivity to greedy feature expansion, and the need for explicit complexity control, descriptor pruning, and more sophisticated search strategies.