Explainable AI-driven design rules for syngas-to-C5+ liquid fuels over cobalt catalysts

The production of liquid fuels from syngas can help alleviate energy supply challenges, support carbon neutrality, and address climate change. However, this process involves considerable complexity due to the interplay of multiple influencing factors, including feedstock characteristics, catalyst properties, and reaction conditions. To facilitate process optimization, we developed a machine learning model to predict CO conversion and C5+ selectivity based on key input descriptors. A dataset of 236 entries was compiled from existing literature, enabling data mining to identify the importance of reaction temperature, reduction degree, and cobalt loading. Analysis revealed that higher C5+ selectivity is achieved at lower temperatures (<240 °C) and moderate cobalt loading (∼20 %). Additionally, it was found that excessively small cobalt particles (<6 nm) negatively impact C5+ selectivity due to increased methane formation and decreased active sites stability at the nanoscale. The proposed framework is entirely data-driven and interpretable, incorporating Permutation Importance (PI), Shapley Additive Explanations (SHAP), and Partial Dependence Plot (PDP), a game theory-based interpretation approach to isolate and analyze the effects of individual and paired descriptors, thereby offering valuable theoretical insights for guiding experimental research.