3D Molecular Generation via Virtual Dynamics

Published: 10 Apr 2024, Last Modified: 12 Apr 2024Accepted by TMLREveryoneRevisionsBibTeX
Abstract: Structure-based drug design, a critical aspect of drug discovery, aims to identify high-affinity molecules for target protein pockets. Traditional virtual screening methods, which involve exhaustive searches within large molecular databases, are inefficient and limited in discovering novel molecules. The pocket-based 3D molecular generation model offers a promising alternative by directly generating molecules with 3D structures and binding positions in the pocket. In this paper, we present VD-Gen, a novel pocket-based 3D molecular generation pipeline. VD-Gen features a series of carefully designed stages to generate fine-grained 3D molecules with binding positions in the pocket cavity end-to-end. Rather than directly generating or sampling atoms with 3D positions in the pocket, VD-Gen randomly initializes multiple virtual particles within the pocket and learns to iteratively move them to approximate the distribution of molecular atoms in 3D space. After the iterative movement, a 3D molecule is extracted and further refined through additional iterative movement, yielding a high-quality 3D molecule with a confidence score. Comprehensive experimental results on pocket-based molecular generation demonstrate that VD-Gen can generate novel 3D molecules that fill the target pocket cavity with high binding affinities, significantly outperforming previous baselines.
Submission Length: Regular submission (no more than 12 pages of main content)
Changes Since Last Submission: We have made the following changes. | change | section | | ---- | ---- | | Revised the writing with more details and clarity and refine the notations | 2 | | Added more case studies | 3.4 | | Added more details to aid reproducibility | 5 | | Added more details about our training dataset and test dataset | 3.1 | | Added Broader Impact Statement | 5 | | Clarified LDDT and pLDDT | 2.1 | | Clarified hyperparameters tuning and added more details about training | Appendix C.1 | | Added an evaluation on the diversity of generated molecules | Appendix C.4 | | Added an in-depth discussion on diffusion models | 4 | | Added details of "Wasserstein distance" approach | 2 | | Clarified how we use the confidence score | 2.1 | | Revised notations throughout the paper for clarity | | | Updated that "SE(3) model" to "SE(3) equivariant model" | | | Checked and updated our citation | | | Added standard deviations to VD-Gen's results | | | Revised the table captions about the standard deviations | | |||
Assigned Action Editor: ~Lei_Li11
Submission Number: 1396