EquiDock is Now Available on Vecura

What is EquiDock?

EquiDock is an SE(3)-equivariant geometric deep learning model that performs end-to-end rigid-body protein-protein docking in a single forward pass. It predicts the bound pose of a ligand protein against a receptor protein by outputting a single rigid-body transformation—a rotation matrix and translation vector—that repositions the ligand into its predicted bound orientation. The model uses an Interacting Equivariant Graph Matching Network that satisfies SE(3) equivariance by design, ensuring predicted poses transform consistently under rotation and translation of inputs.

It helps users rapidly generate structural hypotheses for protein-protein interactions without relying on template structures, candidate-pose sampling, or iterative energy-based refinement. It is especially useful for screening antibody-antigen pairs, guiding mutagenesis studies, and prioritizing candidates for downstream flexible-docking refinement.

What can users do with EquiDock on Vecura?

With EquiDock on Vecura, users can:

• Predict rigid-body docking poses for protein-protein complexes from unbound PDB structures

• Generate complete docked complex structures ready for visualization in PyMOL or Molstar

• Extract rotation and translation transforms to apply in custom analysis pipelines

• Evaluate prediction accuracy against ground-truth structures using RMSD scoring

• Apply optional clash-removal post-processing to reduce atomic overlaps

What the output means

The output provides a complete docked complex PDB file containing both receptor and transformed ligand, the isolated transformed ligand PDB for overlay visualization, the SE(3) rigid-body transform (rotation matrix and translation vector), and optional RMSD scores when ground truth is provided.

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

Why this matters

Protein-protein interactions are fundamental to virtually all biological processes, from signal transduction and immune response to enzyme regulation and cellular organization. Understanding how proteins bind to each other at the structural level is critical for drug discovery, particularly for developing biologics, antibodies, and protein-based therapeutics. Traditional computational docking methods rely on multi-stage pipelines involving extensive sampling, scoring, and energy minimization, which can be computationally expensive and time-consuming.

EquiDock addresses this bottleneck by collapsing the entire docking pipeline into a single learned function that runs in milliseconds. By leveraging SE(3)-equivariant graph neural networks, the model respects the geometric symmetries of 3D space, ensuring physically meaningful predictions regardless of input orientation. This enables researchers to rapidly explore protein interaction landscapes, generate testable hypotheses about binding interfaces, and accelerate the early stages of structure-based drug design. While limited to rigid-body docking, EquiDock provides a powerful first-pass tool that can guide more computationally intensive flexible docking studies and inform experimental design.

• Developed by: Octavian Ganea and colleagues (MIT CSAIL)

• Source: ICLR 2022 paper, official GitHub repository, and pretrained model checkpoints

• Reference: Ganea et al., "Independent SE(3)-Equivariant Models for End-to-End Rigid Protein Docking," ICLR 2022 (arXiv:2111.07786, GitHub: octavian-ganea/equidock_public)