sourced story
2020-2021Primary source · 2 sourcesWell documented

AlphaFold effectively solves protein structure prediction

CASP judges call it a fifty-year grand challenge in biology answered by a neural network

On the timeline · around 2020-2021 · Transformers and the Generative AI WaveTransformers and the Generative AI WaveRecognition and ReasoningAlphaFold effectively solves protein structure prediction20192020202120222023

Quick facts

Institution
Google DeepMind
Competition
CASP14
Result
Median accuracy comparable to experimental methods

What happened

DeepMind's AlphaFold 2 competed in CASP14 (the 14th Critical Assessment of protein Structure Prediction), a biennial blind test in which independent judges compare predicted protein structures against ones solved experimentally but not yet published. AlphaFold produced models for about two-thirds of the CASP14 target proteins with accuracy indistinguishable from experimental methods, measured by a global distance test score above 90 out of 100, a level most competing methods had never reached even once. CASP's assessors, who are not affiliated with DeepMind, stated that AlphaFold had effectively solved the fifty-year-old protein structure prediction problem, predicting a protein's three-dimensional folded shape from its amino acid sequence alone. DeepMind published the full method in Nature in July 2021 and released the AlphaFold source code and, in partnership with EMBL's European Bioinformatics Institute, predicted structures for hundreds of millions of proteins.

Why it matters

Protein structure had been a fundamental bottleneck in biology and drug discovery because determining a single structure experimentally could take years, and AlphaFold's near-experimental accuracy at computational speed gave researchers worldwide immediate access to structures for proteins that had never been solved in a lab, work recognized with the 2024 Nobel Prize in Chemistry.

How we know

The CASP14 organizers' own press release documents the independent judging and the specific accuracy threshold AlphaFold cleared; DeepMind's own account of the project corroborates the same result and its follow-up publication in Nature.

Sources

See something wrong? . Corrections with a source get fixed fastest.

Part of a timelineArtificial Intelligence30 events · From a wartime theory of neurons to machines that write, paint, and fold proteinsView all →