CatEmb Resource Hub: Datasets & Pre-trained Weights

This repository contains the core resources for the CatEmb (Catalyst Embedding) project, which enables the generation of stereoelectronic-aware molecular descriptors directly from 2D graphs. It includes two primary files: model_path.tar.gz: The pre-trained CatEmb model weights. This file allows users to instantly use the CatEmb descriptor for inference without any training, facilitating catalyst similarity analysis and descriptor generation for downstream tasks.dataset.tar.gz: The CatCompDB dataset used to train and evaluate the CatEmb model. This comprehensive dataset of ligands, catalysts, and metal-ligand complexes, with associated 2D structures and processed 3D information, supports the reproduction of our work and serves as a valuable benchmark for future research in data-driven catalyst discovery.Together, these resources provide a complete toolkit for researchers to utilize, validate, and build upon the CatEmb methodology for accelerated catalyst design and discovery.