Streamline Molecular Docking with SigmaDock, Now Available on Vecura

What is SigmaDock?

SigmaDock is an advanced generative AI model designed for protein-ligand molecular docking, utilizing fragment-based SE(3) diffusion to predict how small-molecule ligands bind within a protein pocket. By decomposing ligands into rigid fragments and jointly predicting their translation, rotation, and torsion angles during a reverse diffusion process, the model enables efficient and accurate conformer generation. It helps researchers move beyond traditional search-box methods by performing both re-docking and cross-docking tasks directly from a protein PDB and ligand SDF. It is especially useful for computational chemists and structural biologists working on hit-to-lead optimization and binding pose prediction.

What can users do with SigmaDock on Vecura?

With SigmaDock on Vecura, users can:

• Predict binding poses for ligands in specific protein pockets without manually defining search boxes.

• Perform both re-docking and cross-docking by leveraging reference ligand data to define pocket centers of mass.

• Integrate post-processing workflows, including GNINA-based rescoring (Vina/Vinardo) to rank poses and PoseBusters for structural validation.

• Achieve rapid inference, with the ability to generate docked poses on a single GPU in approximately 12–15 seconds per seed.

What the output means

The output provides an SDF file containing the predicted docked ligand poses, along with comprehensive metadata. This includes seed indices, optional GNINA scoring metrics for rank-ordering, and PoseBusters validity checks ensuring geometric and chemical feasibility (e.g., assessing clashes or valency).

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

Accurate molecular docking is a cornerstone of modern drug discovery, enabling the identification and optimization of potential therapeutics. By automating the prediction of binding poses via diffusion models, SigmaDock addresses many of the challenges associated with the traditional, computationally intensive search-space optimization required by older docking engines.

This integration streamlines the pathway from structure to prediction, reducing the reliance on complex, fragmented local infrastructures and allowing researchers to focus on analyzing binding interactions rather than managing pipeline execution.

• Developed by: Alvaro Prat, Leo Zhang, Charlotte Deane, Yee Whye Teh, Garrett Morris

• Source: GitHub Repository

• Reference: https://arxiv.org/abs/2511.04854