2.2 | Three-dimensional protein structure
As noted earlier, the 2020 CASP round (CASP14) saw a dramatic advance in the accuracy of computational models of single proteins, with the accuracy of many models rivaling that of experiment (4). The few major failures were for oligomeric proteins (for example viral coat components) considered in isolation and some targets with very shallow multiple sequence alignment. Agreement with experiment also tended be lower for the few NMR targets.
For CASP15, interest centered around whether groups other than DeepMind could approach experimental accuracy, whether new approaches would rival AlphaFold2, and whether limitations for shallow sequence alignments would be overcome. DeepMind did not participate in this CASP, but the Elofsson group ran each target through the local standard installation of Alphafold2, providing a baseline.
Figure 1 shows one of the many examples of a protein modeled to high accuracy in this CASP.