2.5 | Macromolecular Ensembles
Previously, CASP has assumed that there is only one relevant structure of a protein to compute, and that structure is represented by a single experimental structure. Increasingly, experimental methods provide more than one structure and computational methods should be able to produce all of these. In this CASP, we introduced an ensembles category with targets that have multiple experimentally observed conformations. As the ensembles evaluation paper discusses (9), there are still very limited experimental data of this type and even when there are data it is not always clear what a computational method should reasonably be expected to deliver. Nevertheless, the CASP15 results do provide preliminary insight into the abilities of computational methods in this area.
There are a number of encouraging results. Of particular note are reproduction of a domain-swap conformational change induced by a single mutation, sampling of three different conformations of an ABC transporter in the presence of different ligands; sampling of kinase substructures observed under different conditions; and sampling of alternative conformations of a small protein found in two different crystal forms. On the other hand, modeling of an RNA folding intermediate proved too difficult, as did different states of a Holliday junction complex. However, these failures may reflect the intrinsic difficulties of the targets as much as sampling deficiencies. The successful methods again used variants of AlphaFold2, with enhanced sampling, including adjustment of the multiple sequence alignment. In this respect, strategies are similar to those used for tertiary structure and protein assembly modeling, though details differ. Many preprints are appearing on improved ensemble modeling methods, so that we expect substantial advances in the next CASP round.