AlphaFold 2 was the AI protein-folding system that achieved a landmark result in the 2020 CASP14 assessment.
Developed by DeepMind, AlphaFold 2 predicted three-dimensional protein structures from amino-acid sequences with accuracy close to experimental methods for many targets. CASP, the Critical Assessment of Structure Prediction, is a biennial blind assessment in which researchers submit predictions for protein structures that have not yet been publicly released.
AlphaFold 2 used a neural-network system that learned relationships among amino-acid sequences, evolutionary information, and three-dimensional structure. Its 2020 performance represented a major advance over earlier computational approaches and helped make structure prediction a practical resource for biology.
AlphaFold 2 did not solve every protein-structure problem. Flexible regions, protein complexes, interactions, and unusual biological states can remain difficult. DeepMind later released a large public database of predicted structures, extending the system’s potential value to researchers in many fields.