Pharmacophore-based design by learning on voxel grids (arxiv.org)

Researchers have developed a novel pharmacophore-based generative model called VoxCap, which utilizes voxel grids to enhance ligand-based drug discovery (LBDD). Traditional LBDD approaches often struggle with scalability and are limited to existing molecular libraries. VoxCap addresses these issues by generating new molecules with high pharmacophore-shape similarities to known binders, thereby facilitating a more diverse candidate generation. This model leverages voxel-based generative modeling to convert voxelized molecular representations into SMILES strings, allowing for efficient and innovative molecular design. The significance of this advancement lies in VoxCap's ability to dramatically accelerate the drug discovery process. It not only outperforms previous methods in generating novel hits but also combines generative design with a fast substructure similarity search, resulting in reduced computational time by orders of magnitude. This approach enables researchers to explore larger molecular libraries that were previously intractable under conventional screening methods. Overall, VoxCap represents a meaningful step forward in the use of AI for drug discovery, promoting the efficient identification of potential drug candidates.