AI Science Tracker

AlphaFold discoveries and experimental evidence

What did AlphaFold predict, and what did researchers confirm through experiments?

All dossiers

What has AlphaFold helped scientists discover?

A study using AlphaFold-Multimer reported experimental support for six previously unreported protein interfaces. CASP14 and the database expansion provide context for the accuracy and availability of structure predictions.

Editorial review 18 Nov 2026. Three catalog examples illustrate different kinds of evidence.

Experiments supported six previously unreported protein interfaces

A study of 62 selected human protein interactions used AlphaFold-Multimer to predict where proteins might contact one another. Follow-up experiments corroborated six previously unreported interfaces. The authors also reported limits in prediction specificity. The experimental support applies to the interfaces tested in this study. Read the protein-interface study.

View the catalog record.

CASP14 tested prediction accuracy

In the blind CASP14 assessment, AlphaFold 2 scored a median 92.4 on the Global Distance Test across targets, according to Google DeepMind. Organizers compared predictions with experimentally determined reference structures. That score describes benchmark performance on the assessed targets. Read the CASP14 announcement.

View the catalog record.

The AlphaFold Database expanded access to predictions

In July 2022, Google DeepMind and EMBL-EBI made more than 200 million predicted structures available. The release greatly expanded the pool of structural models researchers could consult. The number counts predicted structures released in that expansion. Read the 2022 database announcement.

View the catalog record.

Explore the evidence in the catalog

For each study, check what researchers measured and whether the evidence supports a structural prediction, a physical interaction, or a broader biological claim.

Browse the AlphaFold family profile · Read the evidence and selection rules.