AlphaFold3 Is Now Available on Vecura

What is AlphaFold3?

AlphaFold3 is a state-of-the-art AI model developed by Google DeepMind and Isomorphic Labs that predicts the joint all-atom three-dimensional structure of any biomolecular complex directly from sequence. Published in Nature in 2024, it extends the AlphaFold lineage beyond protein-only prediction to a fully unified system that handles proteins, DNA, RNA, small-molecule ligands, ions, and modified residues — all within a single model. It replaces the Evoformer architecture of AlphaFold2 with a more compact Pairformer and an all-atom diffusion module that generates atomic coordinates directly, making it a generative model capable of producing diverse structural hypotheses.

It helps users predict how biomolecules assemble and interact in three-dimensional space — from protein–protein binding to protein–ligand co-folding — without needing separate docking or modeling tools. It is especially useful for understanding molecular interactions in drug discovery, enzyme engineering, and nucleic acid biology where complex multi-component assemblies are the norm.

What can users do with AlphaFold3 on Vecura?

With AlphaFold3 on Vecura, users can:

• Predict protein complex structures — Submit any combination of protein chains and obtain high-confidence all-atom models of their assembled complex, including interface geometry and per-residue confidence scores.
• Model protein–nucleic acid interactions — Co-fold proteins with DNA or RNA strands to predict transcription factor binding, ribonucleoprotein assemblies, and CRISPR guide-target complexes in a single run.
• Co-fold protein–ligand bound states — Provide small-molecule ligands as SMILES strings or PDB CCD codes and generate the bound-state conformation of the protein–ligand complex directly, bypassing traditional docking pose searches.
• Generate diverse structural hypotheses — Run multiple samples per complex using different random seeds to explore conformational diversity, then rank results by a composite score combining pTM, ipTM, and clash penalties.

What the output means

The output provides a ranked set of predicted all-atom structures in mmCIF format, along with per-sample confidence metrics including mean pLDDT (per-residue confidence, 0–100; above 70 is reliable, above 90 is very high), pTM (predicted TM-score for global fold accuracy, 0–1; above 0.5 suggests a correct overall fold), ipTM (interface TM-score for inter-chain accuracy, 0–1), and a ranking score that combines these metrics with a steric clash penalty. A downloadable confidence JSON includes the full per-atom pLDDT array, the pairwise predicted aligned error (PAE) matrix, and per-chain-pair interface metrics for deeper analysis.

This output should be used to support scientific decision making. It does not replace experimental validation. AlphaFold3 predictions are static snapshots — they do not capture conformational dynamics, binding free energies, or allosteric effects. Users should treat high-confidence predictions as strong hypotheses to be confirmed through X-ray crystallography, cryo-EM, NMR, or functional assays.

Why this matters

Understanding how biomolecules interact at the atomic level is one of the most fundamental challenges in biology and medicine. For decades, researchers relied on expensive and time-consuming experimental methods to determine the 3D structures of molecular complexes. AlphaFold2 transformed protein structure prediction in 2020, but it was limited to proteins alone — leaving the vast landscape of protein–DNA, protein–RNA, protein–ligand, and multi-component assemblies largely inaccessible to computational prediction. AlphaFold3 closes this gap by offering a single unified model that predicts the structure of virtually any biomolecular complex, democratizing access to structural insights that were previously out of reach for most labs.

This capability has profound implications for drug discovery, where understanding how a small molecule binds to a target protein — and how that protein interacts with nucleic acids or other proteins — is essential for rational drug design. By enabling researchers to co-fold entire complexes in a single step, AlphaFold3 accelerates the identification of binding interfaces, the design of more selective therapeutics, and the exploration of novel biological mechanisms. Making this model accessible through Vecura means that teams without dedicated GPU clusters or deep learning expertise can now leverage this transformative tool in their daily research workflows.

• Developed by: Google DeepMind / Isomorphic Labs
• Source: Nature 2024 Paper · GitHub Repository · Input Documentation
• Reference: Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature (2024). https://doi.org/10.1038/s41586-024-07487-w