AlphaFold 2 dominates CASP14: a protein-folding breakthrough
In one sentence DeepMind announces AlphaFold 2's CASP14 results: median GDT 92.4 across all targets, above the roughly 90 threshold the organizers consider competitive with experimental methods. For the CASP organizers, the single-protein folding problem is largely solved.
Proteins are the "molecular machines" that make living things work. Knowing the shape a protein folds into is essential to understanding it and designing drugs. For about 50 years biologists had been looking for a way to predict that shape from the amino-acid sequence alone.
DeepMind announces that its AlphaFold 2 system, in the blind international CASP14 competition, predicts the structure of not-yet-published proteins with accuracy often comparable to lab measurements, which can take months or years of work.
It's one of the rare cases where an AI system genuinely moves a big scientific problem, and it opens a new phase for structural biology.
Companies
DeepMind
Tools
AlphaFold 2
Tags
Sources