Cross-Adsorbate Scaling, Interpretable Machine Learning, and Adsorption Trade-Offs across M–N4–C Sites

Metal–nitrogen–carbon sites provide a chemically diverse platform for tuning adsorbate binding, yet the relationships among different C1 adsorption probes remain incompletely resolved. Here, density functional theory and interpretable machine learning are combined across a 35-metal M–N4–C chemical space to examine *COOH, *CO, and *CHO adsorption as complementary probes. The adsorption landscapes are strongly adsorbate dependent, but *COOH and *CHO exhibit a cross-adsorbate scaling relation (R2 = 0.919), consistent with similar metal-dependent stabilization across much of the series. In contrast, relationships involving *CO show larger residual variation. Gradient-boosting models evaluated by nested out-of-fold prediction recover descriptor-based signal for all three targets; *CO is predicted most accurately (R2 = 0.851; MAE = 0.272 eV), whereas *COOH and *CHO show moderate predictability. Permutation importance, direct DFT correlations, and out-of-fold SHAP analysis identify d-electron count as a recurring organizer of *COOH and *CO adsorption and periodic group as a recurring organizer of *CO and *CHO adsorption. Robust regression highlights Os, Ru, and Re as pronounced *COOH–*CO scaling deviations, while Y departs strongly from the tighter *COOH–*CHO relation. Separate Pareto analyses highlight Y, Os, and Ti under different two-objective adsorption trade-offs. The combined analysis distinguishes shared adsorption trends from adsorbate-specific behavior and identifies chemically informative sites for subsequent mechanistic study.