The dataset accommodate all the extra data discussed in:Pisal, P., Krejčí, O. & Rinke, P. Machine learning accelerated descriptor design for catalyst discovery in CO2 to methanol conversion. npj Comput Mater 11, 213 (2025). https://doi.org/10.1038/s41524-025-01664-9 The datased contains four types of data: All the final geometries and energies of adsorbated (*H, *O, *OCHO & *OCH3) and all the 158 single metals and bimetallic alloys on all the surfaces with Miller indices in {-2, -1, ... 2} optimized with Open Catalyst Project (OCP) 20 equiformer_V2 machine-learned force-field model. These are in the geometries_and_energies.zip file organized by the metal/alloys name, with the final geometries and enerigies in a json file, using a json ASE format. All the estimated mean absolute errors (MAE) of predicted adsorption energies for all the considered metals and bimetallic alloys in Estimated_MAEs_metals_bimetallics.csv and xlsx file. The data content is identical, files differs only by a format. All the adsorption energy disctibutions (AEDs) for all the 158 metals/alloys and adsorbates in AEDs_metals_bimetallics.csv and xlsx files. The data content is identical, files differs only by a format. All the statistical information of the adsorption energies for all the 158 metals/alloys and adsorbates in Statistics_AEDs_metals_bimetallics.csv and xlsx files. The data content is identical, files differs only by a format.