RFdiffusion

RFdiffusion is a deep learning model specifically developed for de novo protein design and molecular docking. Based on the architecture of diffusion generative models, it efficiently generates protein structures that meet specific functional requirements, with exceptional performance in protein-protein complex and molecular docking tasks.

The core advantage of RFdiffusion lies in its ability to generate protein structures and model docking complexes by gradually adding noise and guiding its reverse denoising process. This model not only designs entirely new proteins but also predicts the binding structures of molecular ligands (such as small molecules and compounds) with protein targets, providing high-confidence results.

Figure 1: RFdiffusion generates corresponding short peptides based on protein "binding hotspots."