New in Vecura: Streamline Protein Binding Site Prediction with AF2BIND

What is AF2BIND?

AF2BIND is an innovative computational tool that predicts small-molecule binding sites on target proteins by leveraging the internal pairwise representations of AlphaFold2. Instead of relying on traditional 3D-CNNs, it uses a unique "bait" chain strategy to extract structural and co-evolutionary signals that suggest potential binding pockets even in the absence of a known ligand. It helps researchers identify candidate sites for drug discovery by ranking surface residues based on their likelihood of contacting a small molecule. It is especially useful for early-stage drug discovery, guiding fragment screening, site-directed mutagenesis, and pocket-focused docking campaigns.

What can users do with AF2BIND on Vecura?

With AF2BIND on Vecura, users can:

• Identify potential binding sites on a target protein chain without needing a pre-bound ligand.
• Generate per-residue binding probability scores to quantitatively assess which protein regions are most likely to interact with small molecules.
• Obtain an annotated PDB file with B-factors mapped to binding probabilities, allowing for intuitive visualization in tools like Mol* or py3Dmol.
• Generate direct PyMOL selection commands to immediately highlight predicted binding pockets for structural analysis.

What the output means

The output provides per-residue binding probability scores, a correlation matrix regarding chemical ligand properties, and structural files mapped with binding predictions.

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

Why this matters

The identification of druggable pockets is a fundamental challenge in drug discovery, particularly when dealing with "cryptic" sites that only appear in specific conformations or when ligand data is scarce. By repurposing the pre-trained structural knowledge within AlphaFold2, AF2BIND provides a powerful, physics-aware approach to pocket detection without requiring the massive, expensive computational workflows often associated with traditional docking.

This approach bridges the gap between protein structure prediction and functional characterization, offering a high-throughput method to prioritize protein surface areas for experimental investigation.

• Developed by: Gazizov, Lian, Goverde, Ovchinnikov, and Polizzi
• Source: Official GitHub Repository
• Reference: Gazizov et al., "AF2BIND: Predicting ligand-binding sites using the pair representation of AlphaFold2" (2023)