Computing structure-based lipid accessibility of membrane proteins with mp_lipid_acc in RosettaMP

BMC Bioinformatics

Table 3 Prediction accuracies for different parameter sets, showing the stability of the algorithm

	Slice width	Dist cutoff	Shell radius	Angle cutoff	acc	sens	spec
Default	10	10	6	65	91.2	89.4	91.7
	5	10	6	65	89.5	91.4	88.9
	7.5	10	6	65	90.5	90.3	90.6
	10	5	6	65	87.4	92.3	86.0
	10	15	6	65	91.5	81.3	94.5
	10	10	4	65	91.9	82.0	94.8
	10	10	8	65	90.0	91.1	89.7
	10	10	6	50	90.9	89.5	91.4
	10	10	6	80	91.1	87.8	92.1

acc accuracy = (TP + TN)/(TP + TN + FP + FN); sens = TP/(TP + FN); spec = TN/(TN + FP) with TP number of residues predicted as true positives, TN true negatives, FP false positives, FN false negatives
The meaning of the parameters is explained in the Methods section. Our algorithm is stable with respect to smaller slice widths (Fig. 1a) and different angle cutoffs (Fig. 1b). For larger distance cutoffs (Fig. 2d and e) and smaller shell radii (from Fig. 2e to f) the number of predicted true positives decreases while the number of false negatives increases, resulting in a substantial drop in sensitivity (numbers in bold)

ISSN: 1471-2105