Large Language Models Assisted Materials Development: Case of Predictive Analytics for Oxygen Evolution Reaction Catalysts of (Oxy)hydroxides

This study explores the transformative role of artificial intelligence (AI) and machine learning (ML) in materials science, leveraging large language models (LLMs) such as OpenAI’s ChatGPT. Focusing on (oxy)hydroxides as oxygen evolution reaction (OER) catalysts, we demonstrate how LLMs streamline data extraction, significantly reducing reliance on traditional, time-intensive methods. Using few-shot training and strategic prompting, ChatGPT achieved an extraction accuracy of approximately 0.9. The curated data set was then used to predict OER performance via the PyCaret library to evaluate various ML algorithms and a high-accuracy XGBoost regression model with accuracies above 0.9 is subsequently established. Further analysis using SHAP and Python Symbolic Regression (PySR) identified key descriptors-electrochemical double-layer capacitance, transition metal composition, support material, and d-electron count-as critical factors, consistent with established electrochemical principles. Additionally, SHAP’s extreme values for Cu and Zn suggest unconventional catalytic roles, potentially linked to Cu₂O-facilitated NiOOH formation and Zn-induced electronic modulation, demonstrating the power of data-driven analysis in uncovering hidden mechanisms. To enhance literature-based insights, Microsoft’s GraphRAG technology was employed for in-depth chemical information retrieval. Overall, this study introduces an innovative, end-to-end ML framework powered by ChatGPT, promoting broader AI adoption in scientific research and bridging computational intelligence with experimental sciences.