Independent briefing Science & discovery
AlphaFold: five years of impact
How accurate protein-structure predictions changed research workflows—and where experiments still matter.
Independent briefing of Google DeepMind (2025). Read the original for its full evidence, context, and wording.
Start here
Three questions this article answers
What problem did AlphaFold change?
It made high-quality protein-structure predictions available at a scale and speed that experimental structure determination alone could not match.
How are researchers using it?
As a starting point for disease research, molecular biology, conservation, and target exploration, often to prioritize experiments.
Did prediction replace laboratory work?
No. Structures are hypotheses and tools; dynamics, interactions, function, and drug effects still require experimental validation.
Go deeper
Three related questions it doesn’t answer—matched to established papers
How was AlphaFold’s accuracy established?
The Nature paper presents the architecture and benchmark performance behind AlphaFold2.
Paper answering this questionHighly Accurate Protein Structure Prediction with AlphaFoldCan predicted structures support drug discovery?
AlphaFold 3 extends prediction to complexes involving proteins, nucleic acids, ions, and small molecules.
Paper answering this questionAccurate Structure Prediction of Biomolecular Interactions with AlphaFold 3What do static structures fail to reveal?
This article explains why proteins exist as changing ensembles rather than a single fixed shape.
Paper answering this questionProtein Dynamics and Conformational Selection in Signal TransductionThe useful summary
What to take away
AlphaFold’s impact is less about a single dramatic prediction than a change in scientific workflow. Researchers can begin with a plausible structure, compare families of proteins, and decide where expensive experiments will be most informative. Broad access to predicted structures has turned a scarce input into shared infrastructure.
The important boundary is between a model and the molecule. Confidence varies by region and target, and a predicted static structure cannot by itself establish biological function, motion, binding strength, or clinical value. The tool accelerates the route to better questions; it does not waive the need to test them.
Final conclusion
The three answers, brought together
AlphaFold transformed access to protein structures and is used to focus research across many fields. It has not replaced laboratories: prediction guides experiments, while experiments establish dynamics, function, and real-world effect.