AlphaFold

AlphaFold is an artificial intelligence system developed by DeepMind for predicting the three-dimensional structures of proteins from their amino acid sequences. It aims to provide accurate models of protein folding at scale, addressing a long-standing challenge in biology.

The project began with AlphaFold 1, which showed notable accuracy in the protein structure prediction field in 2018. AlphaFold2, introduced in 2020 and published in Nature in 2021, achieved dramatic improvements, delivering near-experimental accuracy for many targets and significantly advancing the field.

AlphaFold uses deep neural networks trained on large repositories of known protein structures. It integrates information from multiple sequence alignments and structural templates to predict inter-residue distances and angles, then assembles these predictions into three-dimensional coordinates. The system outputs per-residue confidence estimates (pLDDT) and a predicted alignment error. AlphaFold-Multimer extends the approach to predict structures of protein complexes.

Impact and availability: In 2021–2022, DeepMind and collaborators released predictions through the AlphaFold Protein Structure Database (hosted by EMBL-EBI), making models available for a substantial portion of the human proteome and many organisms. The resource has been widely used to guide experiments and accelerate research in biology and drug discovery.

Limitations: While highly accurate for many globular proteins, AlphaFold predictions may be unreliable for highly dynamic regions, disordered segments, some multi-domain assemblies, or protein–protein interfaces. Predictions do not directly account for post-translational modifications or environmental factors that influence folding.