Comparison study of two kernel-based learning algorithms for predicting the distance range between antibody interface residues and antigen surface

Abstract

A number of kernel-based machine algorithms have recently been used to study protein–protein interaction in the field of bioinformatics. In this paper we develop a kernel-based machine algorithm called multiple criteria quadratic programming (MCQP) to predict the distance range between antibody interface residues and the antigen surface in antigen–antibody complex. Antibodies bind their antigen using residues, which are part of the hypervariable loops. In this paper we explore the interaction between antibody interface residues and antigen in the study of antibody functions. The distance between the antibody’s interface residue and the antigen surface is one of the antigen–antibody binding characteristics used to observe the details of the antibody–antigen interaction surface. The results predicted by MCQP are compared with those predicted by the support vector machine (SVM). The MCQP algorithm classifies observations into distinct groups based on a number of criteria via a hyperplane. MCQP shows strong advantages for distances of 8 Å and 10 Å. However, SVM gives better results for a distance of 12 Å.

Keywords:

• Multiple criteria quadratic programming
• Support vector machine
• Antibody
• Antigen
• Cross-validation

Acknowledgements

This research was partially supported by a grant from the National Natural Science Foundation of China (grant nos. 70621001, 70531040, 70501030, and 70472074), 973 Project No. 2004CB720103, Ministry of Science and Technology, China, and BHP Billiton Co., Australia.