OpenBind EV-A71 2A is Now Available on Vecura

What is OpenBind EV-A71 2A?

OpenBind EV-A71 2A is a domain-adapted neural network checkpoint of OpenFold3-preview2 specialized for predicting the 3D co-crystal structures of Enterovirus A71 (EV-A71) 2A protease bound to small-molecule ligands. Inheriting the advanced AlphaFold3-style all-atom architecture, it jointly models protein and ligand atoms in a single forward pass. The model was fine-tuned on 925 crystallographic binding events from a public EV-A71 2A fragment-screen dataset to encode a targeted prior over EV-A71 2A binding geometries. It helps users predict high-confidence binding poses and improves scaffold recovery compared to general-purpose structural prediction models. It is especially useful for accelerating compound optimization, analyzing structure-activity relationships, and prioritizing drug-like follow-on compounds for hand-foot-and-mouth disease drug discovery.

What can users do with OpenBind EV-A71 2A on Vecura?

With OpenBind EV-A71 2A on Vecura, users can:

• Predict high-resolution 3D co-crystal structures of the EV-A71 2A protease bound to small-molecule ligands using an intuitive, guided interface.

• Customize the co-folding process by adjusting the number of diffusion samples and model seeds to balance run times and structural pose diversity.

• Automate sequence and template preprocessing with built-in, local mmseqs2 pipelines running against UniRef, ColabFold, and PDB databases.

• Perform target-specific virtual screening of compound candidates, prioritizing them using specialized binding geometry predictions.

What the output means

The output provides a predicted 3D co-crystal structure of the EV-A71 2A protease–ligand complex in PDB or mmCIF format, supplemented by a suite of local and global confidence metrics. These include the predicted Local Distance Difference Test (mean pLDDT) indicating residue-level confidence, the predicted TM-score (ptm) for global fold quality, and the interface predicted TM-score (iptm) which measures confidence specifically at the binding interface.

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

Why this matters

Enterovirus A71 (EV-A71) is the major causative agent of hand-foot-and-mouth disease (HFMD), a high-burden pediatric illness that can cause severe neurological complications and lacks any approved antiviral treatment. The EV-A71 2A protease is an essential enzyme for viral replication, making it a critical drug target. By releasing OpenBind EV-A71 2A, the OpenBind Consortium—co-founded by the University of Oxford and Diamond Light Source—has provided a specialized, open-access AI resource to expedite global therapeutic discovery.

This release represents a significant advancement in target-specific machine learning. While general-purpose models like AlphaFold3 are highly capable, they are often limited by general-domain training data when modeling specific binding pockets. Fine-tuning OpenFold3 on a dense, target-specific dataset demonstrates how combining structural biology with open-source machine learning can deliver tailored models that dramatically improve scaffold recovery and lead optimization, advancing antiviral research for pandemic preparedness.

• Developed by: OpenBind Consortium (co-founded by the University of Oxford and Diamond Light Source, in coordination with the ASAP Discovery Consortium)

• Source: OpenBind Benchmark GitHub Repository and OpenBind Official Website

• Reference: OpenBind Structure–Affinity Data Release (Zenodo)