AlphaFold 2 predicts protein structures with near-experimental accuracy at the CASP14 assessment

Google DeepMind/AlphaFold

At CASP14, the biennial blind test of protein structure prediction, DeepMind's AlphaFold 2 reached a median accuracy score of 92.4 GDT (roughly the share of the protein chain placed correctly); around 90 is considered competitive with lab experiments. The organisers recognised it as a solution to the 50-year-old problem of predicting a protein's 3D shape from its sequence.

• Median GDT score of 92.4 exceeds ~90 threshold for competition with experimental methods.
• Average error ~1.6 Angstroms (0.1 nanometers), comparable to the width of an atom.
• Median score of 87.0 GDT on challenging free-modelling category proteins.
• Recognized as solution to 50-year-old protein structure prediction grand challenge.