DomSVR: domain boundary prediction with support vector regression from sequence information alone

Abstract

Protein domains are structural and fundamental functional units of proteins. The information of protein domain boundaries is helpful in understanding the evolution, structures and functions of proteins, and also plays an important role in protein classification. In this paper, we propose a support vector regression-based method to address the problem of protein domain boundary identification based on novel input profiles extracted from AAindex database. As a result, our method achieves an average sensitivity of approximately 36.5% and an average specificity of approximately 81% for multi-domain protein chains, which is overall better than the performance of published approaches to identify domain boundary. As our method used sequence information alone, our method is simpler and faster.

Fig. 1

Distribution of sequence positions of residues at the center of domain boundaries. Blue dot denotes two-domain chain while red dot stands for protein chain containing more than two domains

Fig. 2

Chain length distributions as observed in the CATH representative set used in this study. Intervals were calculated with a width of 100 residues. The domain frequencies were used to calculate probabilities of predicted domain sizes

Fig. 3

Comparison of raw and smoothing outputs from SVR model for protein chain 1qu6A. The protein chain has 179 residues and contains two domains lined by a domain boundary. The center of the domain boundary is at residue 94. The two types of outputs are normalized to the range [0, 1]. The two square curves denote the two kinds of residue labels. One is true labels describing residues’ states (boundaries/not boundaries); the other is predicted labels

Fig. 4

ROC analysis for mainly alpha proteins with respect to threshold

Fig. 5

ROC analysis for mainly beta proteins with respect to threshold

Fig. 6

ROC analysis for alpha–beta proteins with respect to threshold

Fig. 7

ROC analysis for fewer secondary structures proteins with respect to threshold

Fig. 8

ROC analysis for other proteins with respect to threshold

Fig. 9

Performance comparison based on CASP7 dataset. No left-diagonal striped bars are shown in the right graph for template-based, ab initio, and DomSVR predictors, since the prediction accuracies for three-domain chains are zeros

Fig. 10

Comparison of natural versus predicted domain boundaries for protein chain 1qu6_A. The domain boundary (true or predicted) is shown as space filling grey spheres. a True domain boundary for protein chain1qu6A, b Predicted domain boundary for protein chain1qu6A