Fig. 6: Sampling performance of Mac-Diff and eight competing methods on the allosteric protein AdK and five representative proteins from the Cfold40 test set.
a–f, The 2D plots (top) show 100 sampled structures from each method, with the two axes representing the TM-scores relative to the two experimentally determined conformational states: AdK (a), TAF1 (b), PDI (c), SBDs of OpuA (d), tandem VHS and FYVE domains of HGS-Hrs (e), and surface layer protein (f). The overlays (bottom) highlight the Mac-Diff-generated structures (blue/claret) that achieved the highest TM-scores relative to the two reference structures, with corresponding TM-score and RMSD values also indicated. g, TM-score and RMSD distributions. Data were derived from a sample size of n = 80 independent protein structures, comprising 2 distinct conformations for each of the 40 test proteins. The box plots show the medians as centre lines, the 25th and 75th percentiles as lower and upper quartiles, and 1.5 times the interquartile range as whiskers. The average values of TM-score and RMSD are depicted as black diamonds.
