Electrochemical reduction of carbon dioxide (CO₂) offers a sustainable route to produce valuable chemicals such as formic acid (HCOOH), a promising liquid fuel and industrial feedstock. Previous computational studies suggest that expanding the conjugated ring system of pyridine-based molecular catalysts can lower the activation barrier for CO₂ reduction. In this work, we employ a genetic algorithm (GA) to autonomously explore the vast chemical space of fused heterocyclic catalysts derived from a pyridine core. The GA performs crossover, mutation, and selection on molecular fragments, with fitness defined by activation barriers computed using the single-ended growing string method (GSM). To accelerate the search, the fitness function is evaluated on the fly using machine learning (ML) surrogate models trained on previously computed data, allowing rapid estimation of reaction barriers for new candidates. Through iterative generations, the GA identified novel catalysts containing 7–10 fused aromatic rings with calculated barrier heights as low as 13.5 kcal/mol, significantly lower than pyridine. These results highlight the potential of ML-guided evolutionary algorithms to efficiently discover next-generation electrocatalysts for CO₂ conversion.
