PharmaDiff is a pharmacophore-conditioned diffusion model for 3D molecular generation developed to aid in de novo drug design.
It integrates an atom-based representation of the 3D pharmacophore into the generative process, allowing precise generation of 3D molecular graphs matching predefined pharmacophore hypotheses.
PharmaDiff outperforms ligand-based drug design methods in matching 3D pharmacophore constraints and achieves higher docking scores in structure-based drug design across various proteins.
This model provides a powerful and flexible framework for rational drug design by combining pharmacophore modeling with 3D generative techniques.