AlphaFold2

AlphaFold2 is a protein structure prediction system from DeepMind. Proteins are built from amino-acid chains, and their biological roles depend on the three-dimensional shapes those chains adopt. Predicting that shape from sequence alone is a long-standing problem in biology; AlphaFold2 addresses it with a learned computational model. The model takes an amino-acid sequence and produces a predicted three-dimensional structure, rather than relying only on experimentally determined structures.

The CASP14 result

DeepMind announced on 30 November 2020 that its latest AlphaFold system achieved a median score of 92.4 GDT across all targets in the CASP14 assessment ^[2]. GDT is a scoring measure used to compare predicted structures with reference structures; higher scores indicate closer agreement. DeepMind also said that its CASP14 predictions had an average error of approximately 1.6 Angstroms ^[2]. An Angstrom is a unit of length used at molecular scales, so this figure describes how far, on average, the predicted atomic positions were from the reference positions. DeepMind wrote that a score of around 90 GDT is informally considered to be competitive with results obtained from experimental methods ^[2]. The reported median of 92.4 GDT therefore placed the system in the range that the team described as competitive with experimental results.

The Nature paper

The paper "Highly accurate protein structure prediction with AlphaFold" was published in Nature on 15 July 2021 ^[1]. It states that AlphaFold predicts protein structures with an accuracy competitive with experimental structures in the majority of cases ^[1]. That claim is the central technical result: for many proteins, a computational prediction could be treated as a structural model of comparable quality to an experimental structure. The paper's title and its stated conclusion both emphasize accuracy, which is the property that makes a predicted structure useful for downstream work.

Recognition

On 9 October 2024, the Royal Swedish Academy of Sciences announced that half of the 2024 Nobel Prize in Chemistry goes to Demis Hassabis and John Jumper "for protein structure prediction" ^[3]. The other half went to David Baker "for computational protein design" ^[3]. The award placed protein structure prediction and computational protein design together in the same Nobel Prize in Chemistry. For AlphaFold2, the recognition marked the system's place in a line of work that moved protein structure prediction from a difficult computational problem toward a practical research tool.

Why it matters

AlphaFold2's importance lies in the combination of accuracy and scope reported at CASP14 and in the Nature paper. A median score of 92.4 GDT across all targets, an average error of about 1.6 Angstroms, and accuracy competitive with experimental structures in the majority of cases are the concrete results attached to the system ^[1][2]. These numbers matter because they give researchers a way to judge whether a predicted structure is good enough for a given use. The 2024 Nobel Prize in Chemistry then placed the work in a public scientific context, recognizing protein structure prediction alongside computational protein design ^[3].