Improving the performance of DomainDiscovery of protein domain boundary assignment using inter-domain linker index

Abstract

Background: Knowledge of protein domain boundaries is critical for the characterisation and understanding of protein function. The ability to identify domains without the knowledge of the structure -by using sequence information only -is an essential step in many types of protein analyses. In this present study, we demonstrate that the performance of DomainDiscovery is improved significantly by including the inter-domain linker index value for domain identification from sequence-based information. Improved DomainDiscovery uses a Support Vector Machine (SVM) approach and a unique training dataset built on the principle of consensus among experts in defining domains in protein structure. The SVM was trained using a PSSM (Position Specific Scoring Matrix), secondary structure, solvent accessibility information and inter-domain linker index to detect possible domain boundaries for a target sequence.

DomSVR: Domain Boundary Prediction with Support Vector Regression and Evolutionary Information

“…They are roughly classified into two categories: template-based methods [9][10][11][12][13][14] and ab initio methods [6,[15][16][17][18][19][20][21]. Template based methods use sequence alignment [9], secondary structure alignment [10,11], or other profile alignments to identify domain boundaries.…”

“…Galzitskaya and Melnik predict domain boundaries using side chain entropy of a residue region [16]; DomCut predictor uses amino acid sequence information to predict inter-domain linker regions [17]; Nagarajan and Yona propose a neural network based method that uses the information of multiple sequence alignments analysis, position specific properties of amino acids, and predicted secondary structures [18]; PRODO [19] uses a neural network method with information from position-specific scoring matrix (PSSM) generated by PSI-BLAST [12]; Armidillo converts protein sequences to smoothed numeric profiles using domain linker propensity index (DLI) from amino acids composition and thus predicts domain boundaries [20]; DomainDiscovery applies support vector machines to the detection of domain boundaries with sequence information including a PSSM, secondary structure, solvent accessibility information and inter-domain linker index [21]; DOMpro uses a recursive neural network trained with evolutionary information, solvent accessibility information and secondary structure [11]; Ye et al use a Back-Propagation (BP) neural network method to predict the domain boundary of twodomain proteins with 9 different sequence profiles [22]; recently, Yoo et al propose a new improved general regression network (IGRN) model to predict domain boundaries using a PSSM, secondary structure, solvent accessibility information and interdomain linker index to detect possible domain boundaries [23].…”

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

“…Previous works on the prediction of protein domain boundaries are roughly classified into two categories: template-based methods (Altschul et al 1997; Cheng et al 2006; Gewehr and Zimmer 2006; Marchler-Bauer et al 2007; Marsden et al 2002; Orengo et al 1997) and ab initio methods (Copley et al 2002; Dumontier et al 2005; Galzitskaya and Melnik 2003; George and Heringa 2002b; Nagarajan and Yona 2004; Sikder and Zomaya 2006; Sim et al 2005; Suyama and Ohara 2003). Template-based methods aim to predict domain boundaries using sequence alignment (Marchler-Bauer et al 2007), secondary structure alignment (Cheng et al 2006; Marsden et al 2002), or other profile alignments.…”

“…Among them, CHOPnet addresses some issues in domain annotation with evolutionary information, amino acid composition, and amino acid flexibility (Copley et al 2002); SnapDRAGON predicts domain boundaries using a distance geometry-based folding technique with a 3D domain assignment algorithm (George and Heringa 2002b); Galzitskaya and Melnik (2003) propose a simple approach to identify domain boundaries in proteins using side chain entropy of a residue region; DomCut’s method predicts inter-domain linker regions using amino acid sequence information (Suyama and Ohara 2003); Nagarajan and Yona (2004) propose a neural network-based method to detect domain structure of a protein, which uses the information from multiple sequence alignments analysis, position-specific properties of amino acids, and predicted secondary structures; PRODO (Sim et al 2005) uses a neural network method with information from position-specific scoring matrix (PSSM) generated by PSI-BLAST (Altschul et al 1997); Armadillo aims to predict domain boundaries by converting protein sequences to smoothed numeric profiles based on domain linker propensity index (DLI) from amino acids’ composition (Dumontier et al 2005); Dovidchenko et al (2007) propose a simple and fast method with the use of a minimal number of amino acid sequence alone; DomainDiscovery detects domain boundaries by the use of support vector machines with sequence information including a PSSM, secondary structure, solvent accessibility information and inter-domain linker index (Sikder and Zomaya 2006); DOMpro applies recursive neural network to predict domain boundaries with evolutionary information, solvent evolutionary information, solvent accessibility information, and secondary structure (Cheng et al 2006); Ye et al (2007) present a Back-Propagation (BP) neural network approach to predict the domain boundaries with various property profiles; recently, Yoo et al (2008) develop a new improved general regression network (IGRN) model to detect domain boundaries using a PSSM, secondary structure, information, and inter-domain linker index.…”

Establishing bioinformatics research in the Asia Pacific

“…New methodologies for fast structure retrieval using spectral graph matching [16] and domain boundary prediction using a novel interdomain linker index [17] have been reported. A detailed modelling study of protein dimerization [18] describes how structural resources and methods can be applied to understand biologically functional quaternary structures.…”