Integrating Large Language Models and Machine Learning for Sustainable Electrochemical Upgrading of Biomass-Derived 5-hydroxymethylfurfural to 2,5-bis(hydroxymethyl)furan

The electrochemical hydrogenation of 5-hydroxymethylfurfural (HMF) to 2,5-bis(hydroxymethyl)furan (BHMF) presents a sustainable strategy for biomass valorization, yet selectively upgrading HMF to BHMF with high conversion rate remains challenging. To address this issue, we introduced a computer-assisted framework integrating large language model (LLM), machine learning (ML) and Bayesian optimization to recommend optimal experimental conditions for the HMF-BHMF electrochemical system. We first utilized the multi-modal capacity of GPT-4o to automatically extract literature data via prompt engineering, forming a high-quality dataset comprising 248 experimental entries from 378 publications related to the HMF-BHMF conversion. This LLM-curated dataset was leveraged to train a deep neural network for predicting both conversion and selectivity in a multi-task learning setting. The trained model demonstrated strong predictive performance, achieving R² scores of 0.870 for conversion rate and 0.920 for selectivity. SHAP analysis revealed that electrochemical kinetics and electrode identity were the most influential factors governing conversion and selectivity respectively. Utilizing the machine learning model, Bayesian optimization identified optimized reaction conditions for balancing the trade-off between selectivity and conversion rate. Experimental validation of one such condition yielded 81.70% conversion and 52.08% selectivity, significantly outperforming experimental designs in literature. Our study offers a promising human–machine collaborative approach to accelerating experimental optimization, paving the way for more sustainable electrosynthetic pathways in biomass valorization.