Mean loop prediction accuracy for published methods and the method described in this work (PD2_rmin).

Values for RAPPER were taken from de Bakker et al[64], for LOOPY from Xiang et al[40], Rosetta from Rohl et al[43], FALCm4 from Lee et al[52], FREAD from Choi et al[50], CABS results were estimated from Figure 1 of Jamroz et al[51]. It must be noted that these results do not all come from the same test sets so are not directly comparable. Other methodological differences also make comparison difficult. For example, the Rosetta method repacks all sidechains. The results presented in this table for RAPPER, FALCm4, PD2_rmin and control_rmin are directly comparable as these are predictions based on the Fiser test set [39].