The use of support vector machines to predict protein stability changes caused by point mutation

R.T.J.G. van der Lee* and G. Vriend

Centre for Molecular and Biomolecular Informatics (CMBI), Nijmegen Centre for Molecular Life Sciences (NCMLS), Radboud University Nijmegen Medical centre, 6525 GA 26 Nijmegen

Abstract
Substitution of amino acids provides a way to gain understanding about protein structure and function. Accurate prediction of protein stability changes that result from these mutations is important for designing new proteins. We attempt to develop a support vector machines-based approach to predict stability changes for single site mutations taking into account protein structure properties and general protein knowledge. Furthermore, we give an overview of supervised machine learning techniques and support vector machines in particular. Exploration of the characteristics and limits of support vector machines shows the feasibility for their use in protein stability prediction. We show that our method achieves 69% prediction accuracy on a large dataset of single amino acid mutations, which is lower than the accuracy of other methods. Inspection of protein stability data shows that the data is inconsistent with expectations. Therefore, accurate SVM prediction of protein stability changes using the currently available experimental data seems impossible.

Datasets
P2601
P906

Download pdf