This resource contains two datasets and two lists of candidates for screening in JSON format. Also It contains ZIP folders with all Quantum Espresso Inputs and outputs from which the JSON datasets were obtained. All Quantum Espresso outputs will be later added to Catalysis Hub (https://www.catalysis-hub.org/). The file "QuantumEspresso_versions" is a text file contaning the information of the Quantum Espresso versions employed for obtaining the dataset.

The datasets contain the adsorption energies for surface slabs of a large number of binary intermetallic compounds with different compositions and lattices (for instance, A3B fcc, A3B hpc, AB bcc, etc.). Adsorption energies were computed for different adsorbates (H, O, and OH) on distinct adsorption sites (e.g., fcc AAB, fcc AAA, hcp AAA, hcp AAB, on-top A, and on-top B) and minimum energy surfaces. In addition, different elastic strains (biaxial tension, biaxial compression) were applied to assess their effect on adsorption energies. All calculations were carried out using DFT approximations as implemented in the Open-source software Quantum Espresso. Besides the adsorption energies, the datasets also contain relevant geometric and electronic descriptors (PSI, cell volume, weighted atomic radius, generalized coordination number, weighted electronegativity, weighted first ionization energy, outer electrons, and biaxial strain)  calculated to feed them as features in the training of ML models. The datasets with the tag "scaled" on its name have the descriptors scaled following a MinMax scaling and are given in xlsx format.

The lists for screening contain candidates not included in the dataset for which Random Forest predictions of the Eads were obtained. The lists contain the geometric and electronic descriptors of all screening candidates, as well as the predicted adsorption energy (Eads_RF).

A GitHub repository is linked to this dataset (https://github.com/vvassilevg/HighHydrogenML). The repository contains two Python scripts:

1) Script for creating a dataset from QuantumEspresso outputs, where all relevant descriptors are computed. It outputs a pickle and json files that can be later converted to any other desired format (like xlsx).

2) Script for training a Random Forest model for the prediction of adsorption energies (the datasets with the "scaled" tag must be used for the script to work correctly).

The dataset, ML model and screening have been accepted for publication in Catalysis Science & Technology DOI: DOI:10.1039/D4CY00491D. The accepted Manuscript and the Supplementary information are avilable within this repository.

If you use this dataset or any of the files within this repository, please cite the original publication (10.1039/D4CY00491D) in your work.