Deep Learning-Driven Mechanism Prediction and New Electrocatalyst Generation for Metal-Air Batteries

Abstract In this work, copper-doped Electrocatalyst was synthesized via pyrolysis, and the electrocatalyst was characterized by XRD, FTIR, and XPS, followed by electrochemical testing. Also, a new electrocatalyst for metal-air batteries was designed, and reaction processes are predicted using large language models (LLMs) such as ChemBERT and ChemGPT. To train the models, RDKit was used to process SMILES representations of functional groups, doping techniques, and electrolyte media. While ChemGPT produced new electrocatalysts with an accuracy of 0.83, ChemBERT, which had been optimized to 0.8, was used to predict mechanisms. To speed up catalyst identification, the methodology incorporates reaction pathway predictions.