Fig. 4: Results of the local and global assembly experiments.

a–c Results of the local assembly experiments. a TM-score (n = 27), Pairwise DockQ score (n = 27), and RMSD (n = 27) for all benchmark structures (T = red, O = blue, I = green). Data are represented as boxplots (grey) with the median at the center, 25th percentile at the lower bound, 75th percentile at the upper bound, and whiskers indicating the minimum and maximum values. b Classification of the results using TM-score and Pairwise DockQ score. The outer dashed line delineates successfully predicted structures and the inner black box the highly accurate structures. c Example of an energy landscape (RMSD vs Iscore). d–i results of the global assembly experiments. d TM-score (n = 27), Pairwise DockQ score (n = 27), and RMSD (n = 27) for all benchmark structures (T = red, O = blue, I = green). Data are represented as boxplots (grey) with the median at the center, 25th percentile at the lower bound, 75th percentile at the upper bound, and whiskers indicating the minimum and maximum values. e Classification of the results using TM-score and Pairwise DockQ score. The outer dashed line delineates successfully predicted structures and the inner black box the highly accurate structures. f Example of an energy landscape (RMSD vs Iscore). g Two possible orientations of the oligomer structure in cubic symmetry related by a 180° turn through the \(\psi\) parameter. h Orientation optimization (\(\psi\)) of the subunits within the assembly of the model 5EKW. Models that are started in the wrong orientation are shown at the beginning (generation 0) and at the end (generation 50). Most of the models started in the wrong have learned the correct orientation at the end of the simulation. i: Same simulation as in H but showing snapshots of the distribution of orientation angles (\(\psi\)) as function of the generations. Source data for c, f, h, i are provided in the Source Data file, and the rest in Supplementary Tables 3 and 4.