A Surrogate Machine Learning Model for the Design of Single-Atom Catalyst on Carbon and Porphyrin Supports towards Electrochemistry

We apply the machine learning (ML) tool to calculatethe Gibbsfree energy (Delta G) of reaction intermediatesrapidly and accurately as a guide for designing porphyrin- and graphene-supportedsingle-atom catalysts (SACs) toward electrochemical reactions. Basedon the 2105 DFT calculation data from the literature, we trained asupport vector machine (SVR) algorithm. The hyperparameters were optimizedusing Bayesian optimization along with 10-fold cross-validation toavoid overfitting. Based on the Shapley Additive exPlanation (SHAP)and permutation methods, the feature importance analysis suggeststhat the most important parameters are the number of pyridinic nitrogen(Npy), the number of d electrons (theta(d)), and the numberof valence electrons of reaction intermediates. Inspired by this featureimportance analysis and the Pearson correlation coefficient, we founda linear dependent, simple, and general descriptor (phi) to describe Delta G of reaction intermediates (e.g., Delta G (OH*) = 0.020 phi -2.190). Using the trained SVR algorithm, Delta G (OH*), Delta G (O*), Delta G (OOH*), Delta G (OO*), Delta G (H*), Delta G (COOH*), Delta G (CO*), and Delta G (N2*) intermediates are predicted for the oxygenreduction reaction (ORR), the oxygen evolution reaction (OER), thehydrogen evolution reaction (HER), and the CO2 reductionreaction (CO2RR). The SVR model predicts an ORR overpotentialof 0.51 V and an HER overpotential of 0.22 V for FeN4-SAC. Moreover,we used the SVR algorithm for high-throughput screening of SACs, suggestingnew SACs with low ORR overpotentials. This strategy provides a data-drivencatalyst design method that significantly reduces the costs of DFTcalculations while providing the means for designing SACs for electrocatalysisand beyond.