Figure 2: Percentage of targets modeled to three Cα RMSD accuracy thresholds (1, 2 and 3Å) in the four most recent CASP experiments. The fraction of high accuracy structures increased dramatically from 2016 (CASP12) to 2018 (CASP13) because of the introduction of effective deep learning methods and again from 2018 to 2020 (CASP14) with the introduction of the AlphaFold2 deep learning method. Increases from 2020 to 2022 (CASP15) are more modest likely because in CASP14 many computed structures were already within experimental uncertainty, so there is not much room for further improvement. RMSD (root mean square deviation) includes all common residues in the experimental and computed structures.
Figure 3 shows progress over all the CASP experiments using the more robust GDT_TS measure of backbone agreement, and is an update of equivalent figures shown in earlier CASP overview papers. As in the previous 2020 CASP14 (blue line), and consistent with the RMSD result in figure 2, the majority of best models (black line and open circles) approach experimental accuracy (approximately 90% on the GDT_TS scale), and overall best performance is very similar. There is only a small fall-off with the extent to which homology modeling can be utilized (X axis difficulty scale). This result is consistent with expectation, since once experimental accuracy is reached there is no way of measuring further improvement. Of note, best server performance (dotted line) is similar to that returned by groups where human intervention was possible, showing that most procedures could be fully automated.