In November 2020, DeepMind’s AlphaFold achieved a breakthrough at CASP14, the biennial blind contest for predicting protein 3D structure from amino acid sequence. The protein-folding problem had stood as a grand challenge for roughly fifty years. DeepMind’s own announcement, dated November 30, 2020, reports a median score of 92.4 GDT across all targets.
The peer-reviewed account followed in Nature on July 15, 2021, in “Highly accurate protein structure prediction with AlphaFold” by John Jumper, Richard Evans, and colleagues. The paper reports a median backbone accuracy of 0.96 Angstroms, compared with 2.8 Angstroms for the next-best method, and states that AlphaFold can regularly predict structures with atomic accuracy even when no similar structure is known.
Proteins are the molecular machines of biology, and knowing their shape is central to understanding disease and designing drugs. Determining a structure in the lab can take months or years; AlphaFold made accurate prediction fast and cheap.
This was a landmark demonstration that deep learning could solve a hard scientific problem of real-world consequence, not just play games or process text. It reshaped expectations for AI as a tool for discovery across the sciences.