Fig. 3: Comparison of predictions with ideal ion mobility (IM) data to those of AlphaFold (AF) and RF (RosettaFold).
Predictions, for the ideal dataset, using the IM score function were compared to that of Rosetta (RS), AFwith_templates, and RFwith_templates. a Violin distributions (n = 60 biologically independent samples over 4 independent modelling approaches), of (i) root mean square deviation (RMSD) and (ii) template modelling score (TM-Score) of the predicted structure using the RS, IM, AFwith_templates and RFwith_templates score functions. For protein structure predictions with methods shown in a the mean and the standard error of mean in (i) are 4.46 ± 0.74 Å, 3.72 ± 0.65 Å, 3.43 ± 0.77 Å, and 4.37 ± 0.80 Å respectively. Similarly the mean and the standard error of mean in (ii) are 0.84 ± 0.03, 0.86 ± 0.02, 0.92 ± 0.02, and 0.88 ± 0.02 respectively. For the violin distributions in a (i) and a (ii) the white dots represent the median. The black bar in the center of the distribution is the interquartile range (IQR). The black stretched line extends from the “first quartile −1.5 IQR” to the “third quartile + 1.5 IQR”. Values beyond this range are considered outliers. b Comparison of predicted structures with AFwith_templates (cyan), RFwith_templates (red) and IM (purple) to their native structures (grey) for the ideal dataset. c High normalized absolute difference in collision cross section of the predicted structure and the native structure (ΔCCS divided by sequence length) for structures predicted with AFwith_templates (cyan) and RFwith_templates (red) generally corresponded to structures with low TM-Score as seen for the ideal dataset. Source data are provided as a Source Data file.
