Conformal Active Learning Aided Screening of Ligand Protected Cu-Nanoclusters for CO2 Reduction Reactions

Abstract

In this study, we have proposed a conformal active learning (CAL) method to screen the ligands protected atomically precise Cu-nanoclusters for CO2 reduction reaction. Here, we have investigated the roles of the core metals and the protecting ligands towards the product selectivity. We demonstrate here that such unique machine learning model can accurately accelerate the screening of nanoclusters where prediction uncertainty for unknown datasets can be calculated with rigorous coverage guarantee resulting reduced error and uncertainty (mean prediction interval) in the prediction of overall datasets. Thus, this model effectively balances the uncertainty, coverage guarantee and the prediction error for all the considered descriptors important for a catalytic reaction. Moreover, through feature importance analysis, we have determined the major influences from the considered features. Finally, screening through all the predicted results of each of the descriptors and putting some optimum criteria, we can screen out the best selective catalysts for C1 product formation. Such CAL-based method will certainly open a new direction in the field of high throughput screening of catalysts.