GeoLinker: Pre-trained Models and Datasets for 3D Molecular Linker Design

This repository contains the official pre-trained model weights and processed datasets for "GeoLinker: An Equivariant Diffusion Model for 3D Molecular Linker Design". Included Files:1. geolinker_best_unified.pt: The unified EGNN model checkpoint trained on the ZINC dataset across all four conditioning modes (Base, Anchor, Sized, Both).2. geolinker_datasets.zip: Processed 3D point cloud datasets (.pt format) for ZINC, CASF-2016, and GEOM-drugs, including PAINS filters and training linker SMILES required for novelty evaluation. Code to run these models can be found on our official GitHub repository.