A Machine Learning-Guided Study of Structure-Reactivity Relationships in Diels-Alder Cycloadditions

The Diels-Alder cycloaddition is a cornerstone transformation in organic synthesis and has been extensively studied in both experimental and theoretical contexts. In this work, we present a complementary computational approach that combines density functional theory (DFT) and machine learning to further elucidate the role of steric and electronic effects in determining the reactivity and activation barriers. A diverse dataset of 1000 uncatalyzed hydrocarbon Diels-Alder reactions was used to train predictive models that relate activation energies to chemically meaningful molecular descriptors. SHAP analysis of the machine learning models highlights the dominant influence of steric effects, particularly those associated with substituent volume at the internal diene carbons, which can impose conformational strain and lead to significantly elevated barriers. In contrast, substituents at the terminal positions have a more limited impact. We introduced the minimum energy gap between LUMO_diene-HOMO_dienophile and LUMO_dienophile-HOMO_diene as a key predictive descriptor. This feature shows a strong correlation with the activation energy across the dataset, although steric interactions can lead to notable deviations from the overall trend. The resulting models provide insights for rationalizing selectivity and designing more efficient cycloadditions based on steric and electronic complementarity.