AlphaFold

Assembled from 40 blocks · 3 evidence · 45 related

1. Story
2. The protein structure prediction system built at Google DeepMind, and the strongest single instance of machine learning demonstrably changing a scientific practice. The problem it addresses — inferring a protein's three-dimensional fold from its amino-acid sequence — had been open for roughly fifty years and was the subject of a standing blind experiment, CASP, precisely because the field could not agree on whose method worked. At CASP14 in late 2020 AlphaFold2 produced predictions with a median domain GDT_TS reported at 92.4, a level of accuracy the assessors described as competitive with experimental determination for most targets; the method was published in Nature in July 2021 (596, 583–589). What makes this different from most capability claims is that the test was blind, externally run and not chosen by the claimant, and that the result was then used: predicted structures became a routine first step in structural biology rather than a curiosity. Its limits are equally documented and are not marginal — see the same records.
3. Knowledge
4. Deep learning
5. AlphaFold
6. John Jumper
7. Google DeepMind
8. Connections
9. CASP — Critical Assessment of Structure Prediction
10. Google DeepMind
11. John Jumper
12. Deep learning
13. AlphaFold Protein Structure Database
14. AlphaFold2 is assessed blind at CASP14
15. AlphaFold2 is published and the database opens
16. The AlphaFold database expands to over 200 million structures
17. The Nobel Prize in Chemistry recognises protein structure prediction
18. AlphaFold
19. GNoME and the materials-discovery claim
20. London
21. AlphaGo defeats Lee Sedol in Seoul
22. AlphaFold2 is assessed blind at CASP14
23. GNoME is announced as 2.2 million new crystal structures
24. GenCast is published with a like-for-like comparison against an operational ensemble
25. AlphaFold
26. AlphaFold2 is assessed blind at CASP14
27. AlphaFold2 is published and the database opens
28. The Nobel Prize in Chemistry recognises protein structure prediction
29. The AI accelerator
30. ImageNet
31. The transformer architecture
32. AlphaFold
33. Machine-learned weather forecasting
34. Machine learning in medical imaging
35. A deep convolutional network wins the ImageNet challenge
36. AlphaGo defeats Lee Sedol in Seoul
37. Evidence
38. AlphaFold2 predicts protein structure from sequence at accuracy competitive with experimental determination for most targets, as assessed blind at CASP14 in 2020, where predictions are reported to have achieved a median domain GDT_TS of 92.4 including on free-modelling targets. V55 verification basis: the paper was NOT fetched — nature.com is blocked to this session — but search results carried the title, journal, volume 596 and pages 583–589 and attributed them to this paper, and carried the CASP14 accuracy characterisation. The 92.4 median GDT_TS figure is attributed in retrieved summaries to the CASP14 assessment literature rather than to this paper, and a curator should confirm which document reports it before it is quoted as the paper's own statistic.
39. The database opened in July 2021 with the human proteome and model organisms, expanded on 28 July 2022 to more than 200 million predicted structures covering effectively the whole of UniProt, and is reported in a 2024 database paper as covering over 214 million sequences; the FAQ states that AlphaFold has not been validated for predicting the effects of destabilising point mutations and that its output is a single conformation rather than a sample of a conformational ensemble. V55 verification basis: no EMBL-EBI page was fetched; search results carried the July 2022 two-hundredfold expansion, the 214-million figure and the paper title with attribution to EMBL-EBI, and carried the mutation and single-conformation limitations with attribution to the AlphaFold DB FAQ. The reported size at the July 2021 launch differs between summaries (figures near 300,000 and near 350,000 both appear) and is therefore not asserted. The adoption figures — over three million researchers in more than 190 countries by late 2025, over a million of them in low- and middle-income countries, and an independently analysed rise of over 40% in users' submissions of novel experimental structures — reached this record through secondary reporting of DeepMind and EMBL-EBI communications and require confirmation.
40. The 2024 Nobel Prize in Chemistry was divided, one half to David Baker "for computational protein design" and the other half jointly to Demis Hassabis and John Jumper "for protein structure prediction"; the citation materials describe the prediction of protein structure from amino-acid sequence as a problem chemists had wrestled with for over fifty years. V55 verification basis: nobelprize.org was not fetched, but search results carried the exact citation wording and the division of the prize and attributed them to the Academy's own pages.