Discovering Electron-Sponge Behavior at Organic-Metal Interfaces for CO₂ Electroreduction via Machine Learning

Molecular regulation at organic-metal interfaces is crucial for C─C coupling in CO₂ electroreduction, directly influencing the formation of multi-carbon (C₂₊) products. However, the non-linear interplay of electronic, spatial, and topological molecular descriptors has hindered the establishment of predictive quantitative structure-activity relationships (QSAR), limiting mechanistic insight. Herein, we employed an interpretable machine learning (ML)-QSAR framework to link molecular features with the C─C coupling free energy barrier (ΔG‡) on Cu surfaces, uncovering the dominant role of interfacial "electron-sponge" behavior. Mechanistically, the modifier molecule donates electrons to Cu, which subsequently redistributes them to *CO/*CHO intermediates and the molecule itself, while also directly stabilizing the intermediates. Shapley Additive Explanations (SHAP) analysis identifies key electronic descriptors, including low minimal local electron affinity (LEA_min), narrow HOMO-LUMO gap and elevated HOMO energy. These descriptors govern the electron-sponge mechanism, facilitating the reduction of ΔG‡. As a representative molecule, 3,4-diaminofurazan (DAF), selected from a library of 5,304 graph-theory-derived compounds, incorporates electron-donating and back-donating amino and furan-azole motifs. Experimental validation shows a 1.8-fold increase in C₂₊ Faradaic efficiency, from 42% to 77%, confirming the QSAR framework's effectiveness. This descriptor-driven approach was further extended to Au and Ag systems, providing a scalable pathway for designing next-generation electrocatalysts.