Google DeepMind and EMBL-EBI
Verified evidenceContinuous decisioningResearchers can use precomputed structure predictions, then decide what experimental validation is required.
Engineering design · Global/open scientific resource
Collections: Embodied work
Executive brief
The transformation is infrastructural: a once-scarce intermediate artifact became available on demand, shifting scientists from acquisition toward interpretation and validation.
AI value · Protein sequences with openly accessible predicted structures
VerifiedMore than 214 million predicted structures, covering nearly the full UniProt database
Coverage is not equivalent to experimental accuracy or research impact; predictions have confidence limits and do not replace validation for many uses.
Before
Search the much smaller experimental structure corpus or initiate a bespoke structure-determination project. → Wait for a structure before pursuing structure-informed hypotheses.
After
Serves a precomputed predicted structure and confidence metrics for a queried protein sequence. → Uses predictions to form hypotheses, design experiments, or solve structures, validating consequential uses.
Human boundary
AlphaFold supplies a prediction; scientists decide whether confidence is sufficient and what experimental validation is required.
Why it matters
Researchers can use precomputed structure predictions, then decide what experimental validation is required.
Before
How the work ran before the change.
Structural biologist
Search the much smaller experimental structure corpus or initiate a bespoke structure-determination project.
ControlExperimental feasibility
Research team
Wait for a structure before pursuing structure-informed hypotheses.
ControlScientific judgment
What changed
Researchers can use precomputed structure predictions, then decide what experimental validation is required.
Decision rightSelection moves from a fixed rule to the model
After
How the same work runs now.
AlphaFold DB
Serves a precomputed predicted structure and confidence metrics for a queried protein sequence.
ControlPrediction confidence and coverage
Researcher
Uses predictions to form hypotheses, design experiments, or solve structures, validating consequential uses.
ControlHuman scientific validation
Exception path
Low-confidence regions, multimers, nonstandard sequences, and consequential claims require alternative methods or experiments.
Work removed
Decision authority
AlphaFold supplies a prediction; scientists decide whether confidence is sufficient and what experimental validation is required.
Before
Structural biologist
Search the much smaller experimental structure corpus or initiate a bespoke structure-determination project.
Control: Experimental feasibility
Research team
Wait for a structure before pursuing structure-informed hypotheses.
Control: Scientific judgment
After
AlphaFold DB
Serves a precomputed predicted structure and confidence metrics for a queried protein sequence.
Control: Prediction confidence and coverage
Researcher
Uses predictions to form hypotheses, design experiments, or solve structures, validating consequential uses.
Control: Human scientific validation
Work that left the path
Human role before
Researchers spent scarce experimental or computational capacity obtaining initial structural information.
Human role after
Researchers start from a prediction, interpret confidence, and focus experiments on validation and downstream questions.
AI roleProtein-fold prediction precomputed at database scale and exposed through web, API, and bulk access.
Approximately 200,000 PDB structures representing about 60,000 unique sequences→More than 214 million predicted structures, covering nearly the full UniProt database
Database expansion from 2021 through 2024 · Global open database used by researchers in 190 countries
Coverage is not equivalent to experimental accuracy or research impact; predictions have confidence limits and do not replace validation for many uses.
Anti-pattern
Treating a predicted structure as experimentally confirmed or counting database entries as discoveries.
Questions
Portability conditions
Reputation risk
low
Evidence and authority
Current · updated
1 peer reviewed, 1 primary; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID 027158736e730ddc
Google Research Flood Hub · Continuous decisioning
A global model can transfer hydrology into ungauged basins.
European Centre for Medium-Range Weather Forecasts · Continuous decisioning
An ML forecast can run as a supported production system alongside retained physics-based forecasting.
Mayo Clinic primary-care practices · Continuous decisioning
A routine ECG can trigger evaluation for asymptomatic low ejection fraction before specialist suspicion.
Google · Continuous decisioning
Novel breakages need not wait for a comprehensive post-submit test cycle.