Predicting changes in protein thermostability brought about by single- or multi-site mutations

BMC Bioinformatics

Table 3 Performance of Prethermut on the M-dataset with different ranges of absolute ΔΔG

Method^a	Range of absolute ΔΔG	m^b	MCC	Q2 (%)	Sensitivity (%)	Specificity (%)	r
RF	[0, 1)	1466	0.33	66.8	68.9	65.5	0.39
RF	[1, 2)	873	0.57	84.0	78.7	85.2	0.56
RF	[2, 3)	509	0.66	91.0	88.1	91.3	0.69
RF	[3, 14)	518	0.77	94.8	87.9	95.7	0.72
SVM	[0, 1)	1466	0.28	68.3	36.9	87.1	0.31
SVM	[1, 2)	873	0.52	86.3	49.7	95.0	0.55
SVM	[2, 3)	509	0.64	93.3	57.6	98.0	0.65
SVM	[3, 14)	1466	0.62	93.4	44.8	99.6	0.63

All results were obtained by a 10-fold cross validation on the M-dataset. See Methods for definitions of overall accuracy (Q2), Matthews correlation coefficient (MCC), sensitivity, specificity, and Pearson correlation coefficient (r). ^aThe number of trees in the random forests (RF) method is 10000; the parameters for the support vector machine (SVM) method are gamma (g) = 2, cost (c) = 8, and the weight for the positive samples (w) = 3. ^bm is the number of mutant proteins in the M-dataset that have the same range of absolute ΔΔG.

ISSN: 1471-2105