Supplementary Data: Machine Learning Framework for Stability Evaluation of Dual-Atom Catalysts

ML-DACs is a machine learning framework for evaluating the stability of Dual-Atom Catalysts (DACs) against aggregation. The framework automates the workflow from data preprocessing to model training and evaluation, enabling efficient prediction of DAC stability to support the rapid screening and design of new catalyst materials. This repository contains the stability descriptors and DFT-simulated adsorption energies of DACs on N-doped carbon used within the ML-DACs framework. Pipeline available in the github.com/mminotaki/ml_dacs Summary dataset (final selected descriptors + target, for inspection): ml_dacs_supplementary_data.csv — the dataset in CSV format. ml_dacs_supplementary_data.pkl — pickle version of the same dataset. Model-input files (place in data/external/dacs_energies_out/; start from 30_rfr_training.ipynb). Edft_balanced_df.pkl is the primary dataset used for the reported Random Forest results. The remaining files are per-configuration subsets used to inspect performance across cavity/coordination groups: Edft_balanced_df.pkl — primary model input Edacs_dft.pkl Edft_din6_s_df.pkl, Edft_din6_as_df.pkl, Edft_din6_df.pkl Edft_din4_x2_df.pkl Edft_din6_s_din4_x2_df.pkl, Edft_din6_as_din4_x2_df.pkl Raw DFT outputs and optimized geometries are available on ioChem-BD; the data-preparation notebooks (01–02) require that raw data, while the ML results reproduce directly from the files above.