Molecules generated by UniLingo3DMol: a unified language model bridging de novo and fragment-based 3D molecule design

This repository contains the essential resources for UniLingo3DMol, a unified language model designed to bridge de novo and fragment-based 3D molecular design. We provide all generated molecules across the following three benchmark datasets: DUD-E (de novo design: dude-denovo.zip) CrossDocked2020 (de novo design: crossdocked_denovo.zip) Custom DUD-E (fragment-based design: dude-fragment-based.zip) We provide all scripts for obtaining evaluation results in our study: DUD-E (de novo design: dude_denovo_set_metrics_analysis.ipynb) CrossDocked2020 (de novo design: crossdocked_denovo_set_metrics_analysis.ipynb) Custom DUD-E (fragment-based design: dude_fragment_retained_sets_metrics_analysis.ipynb) Additionally, we include the results of two rounds of molecular filtering for wet-lab experiments (cblb_inhibitor_discovery.zip), with the molecules from each round organized into separate subdirectories.