Autoregressive fragment-based diffusion for pocket-aware ligand design
This work addresses ligand design for drug discovery, presenting an incremental method for pocket-aware molecular generation.
The authors tackled the problem of generating 3D molecular structures conditioned on protein targets by introducing AutoFragDiff, a fragment-based autoregressive diffusion model, which improves local geometry while maintaining high predicted binding affinity.
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.