AlphaFold 2 predicted protein three-dimensional structures in the 2020 CASP14 assessment.
Developed by DeepMind, AlphaFold 2 used deep-learning methods to infer a protein’s folded structure from its amino-acid sequence. In the fourteenth Critical Assessment of protein Structure Prediction, commonly called CASP14, it achieved a major advance over competing computational methods on many targets.
The achievement mattered because proteins’ shapes strongly influence their biological functions, while experimentally determining structures can be slow and expensive. AlphaFold 2 did not solve every protein-structure problem, and predictions can be less reliable for flexible regions, complexes, or unusual biological situations. Researchers therefore treat confidence estimates and experimental evidence as important.
AlphaFold 2 is distinct from AlphaGo, DeepMind’s Go-playing system. The “2” also matters: the 2020 CASP14 result is specifically associated with the second major AlphaFold system, not simply any software carrying the AlphaFold name.