Advanced 3D Molecular Design with PMDM: Now Available on Vecura

What is PMDM?

PMDM (Pocket-aware Molecular Diffusion Model) is a structure-based drug design tool that leverages a dual equivariant graph neural network to generate 3D bioactive small molecules conditioned on a specific protein binding pocket. By utilizing separate global-graph and local-graph denoisers, it effectively captures both long-range protein-ligand interactions and local bonding geometry. This model helps medicinal chemists and computational researchers accelerate lead discovery and optimization by iteratively denoising ligand point clouds within a defined target site.

What can users do with PMDM on Vecura?

With PMDM on Vecura, users can:

• Extract Binding Pockets: Easily prepare protein structures for generation by isolating the relevant binding site from a holo protein-ligand complex PDB.

• Perform De Novo Generation: Generate novel 3D small molecules tailored specifically to a target pocket based on a desired heavy-atom count.

• Execute Fragment Growing: Expand a seed fragment into a larger, optimized lead candidate by generating new atoms around specific anchors within the binding site.

• Design Linkers: Replace masked regions in a scaffold with novel linkers, enabling precise modification of existing molecules to improve binding or pharmacological properties.

What the output means

The output provides 3D molecular structures in SDF format, corresponding canonical SMILES strings, and essential chemoinformatics metrics including QED (Quantitative Estimate of Drug-likeness), SA (Synthetic Accessibility), and Lipinski rule-of-five compliance.

This output should be used to support scientific decision-making in drug discovery. It does not replace experimental validation or downstream physical affinity testing.

Why this matters

The ability to generate 3D-consistent, pocket-conditioned molecules is a significant bottleneck in computational drug design. By providing a scalable, cloud-based implementation of PMDM, researchers can move beyond traditional virtual screening libraries and explore novel chemical space tailored to the unique geometric and electronic environment of a target receptor.

This capability is particularly valuable for tackling challenging protein targets where standard libraries may lack sufficient binding diversity, ultimately streamlining the transition from computational hypothesis to experimental testing.

• Developed by: Layne Huang et al.

• Source: GitHub repository

• Reference: Huang, L. et al. "Pocket-aware molecular diffusion model for structure-based drug design." Nature Communications (2024).