Prodepth: Predict Residue Depth by Support Vector Regression Approach from Protein Sequences Only

• DOI: 10.1371/journal.pone.0007072
• Publication Date: 2009-09-16T23:13:37Z
• Authors: Jiangning Song; Hao Tan; Khalid Mahmood; Ruby H. P. Law; Ashley M. Buckle; Geoffrey I. Webb; Tatsuya Akutsu; James C. Whisstock
• Keywords: Accessible surface area; Protein sequencing; Sequence (biology); Root mean square

Residue depth (RD) is a solvent exposure measure that complements the information provided by conventional accessible surface area (ASA) and describes to what extent residue buried in protein structure space. Previous studies have established RD correlated with several properties, such as stability, conservation amino acid types. Accurate prediction of has many potentially important applications field structural bioinformatics, for example, facilitating identification functionally residues, or residues folding nucleus, enzyme active sites from sequence information. In this work, we introduce an efficient approach uses support vector regression quantify relationship between sequence. We systematically investigated eight different encoding schemes including both local global characteristics examined their respective performances. For objective evaluation our approach, used 5-fold cross-validation assess accuracies showed overall best performance could be achieved correlation coefficient (CC) 0.71 observed predicted values root mean square error (RMSE) 1.74, after incorporating relevant multiple features. The results suggest reliably solely primary sequences: environments are major determinants, while features influence marginally. highlight two examples comparison order illustrate applicability approach. also discuss potential implications new parameter homology modeling. This method might prove powerful tool analysis.