Prediction of hot spot residues at protein-protein interfaces by combining machine learning and energy-based methods

Table 1 Summary of results for the binary classification problem

	Model	Precision	Recall	F1 score	MCC
(a)	SVM	0.53 ± 0.03	0.63 ± 0.04	0.58 ± 0.02	0.44 ± 0.03
	TSVM	0.56 ± 0.03	0.65 ± 0.03	0.60 ± 0.02	0.47 ± 0.02
	GP	0.59	0.32	0.41	0.33
	Robetta₂	0.52	0.47	0.49	0.35
	Robetta_1.8	0.53	0.52	0.52	0.38
	Robetta₁	0.39	0.75	0.52	0.34
(b)	SVM	0.64 ± 0.03	0.79 ± 0.05	0.71 ± 0.01	0.40 ± 0.03
	Robetta₁	0.69	0.65	0.67	0.38

In (a) experimental hot spots are defined as those residues for which ΔΔG ≥ 2 kcal/mol, in (b) a threshold of 1 kcal/mol is used. SVM: Support Vector Machine, TSVM: Transductive SVM, GP: Gaussian Processes, : Robetta scores (estimated ΔΔG values) which are greater than x_thare considered as predicted hot spots. MCC is the Matthews correlation coefficient (see Methods section for definition of the various performance measures). Results for SVM, TSVM and GP have been obtained with a 16-fold cross-validation scheme. Results for Robetta have been retrieved from the server at http://robetta.bakerlab.org/.

ISSN: 1471-2105