In this work, we introduce AutoFragDiff, a fragment-based autoregressive diffusion model for generating 3D molecular structures conditioned on target protein structures. We employ geometric vector perceptrons to predict atom types and spatial coordinates of new molecular fragments conditioned on molecular scaffolds and protein pockets. Our approach improves the local geometry of the resulting 3D molecules while maintaining high predicted binding affinity to protein targets. The model can also perform scaffold extension from user-provided starting molecular scaffold.
@article{arxiv.2401.05370,
title = {Autoregressive fragment-based diffusion for pocket-aware ligand design},
author = {Mahdi Ghorbani and Leo Gendelev and Paul Beroza and Michael J. Keiser},
journal= {arXiv preprint arXiv:2401.05370},
year = {2024}
}
Comments
Accepted, NeurIPS 2023 Generative AI and Biology Workshop. OpenReview: https://openreview.net/forum?id=E3HN48zjam