Conceptual <scp>DFT</scp> Meets Machine Learning: A New Route to Enhanced Diels–Alder Reactivity

Conceptual Density Functional Theory (CDFT) offers a rigorous framework for understanding chemical reactivity through energy-response descriptors. The global electrophilicity index (ω) is particularly useful for rationalizing trends in reactions such as Diels-Alder (DA) cycloadditions. In this work, we present a deep-learning model predicting ω from randomized Coulomb matrices derived from force-field geometries. This enables high-throughput screening with accuracy comparable to DFT but at a fraction of the cost. Validation shows prediction errors below 0.1 eV for nucleophiles and ~0.3 eV for less abundant electrophiles from our PubChem and GDB13 datasets. Diels-Alder activation barriers of a curated set of dienophiles with furan and fulvene correlated well with ω, especially when considering the transition-state interaction energy. Of value for applications in bioconjugation and self-healing polymers, several candidates were identified as more reactive than maleimide (ω = 1.63 eV). Machine-learning models targeting CDFT descriptors thus provide scalable, interpretable tools for automated reaction-space exploration.