Prediction of EF-hand calcium-binding proteins and identification of calcium-binding regions using machine learning techniques

Predicting calcium binding proteins and identifying calcium-binding sites is an important problem in the field of proteomics. Most of the currently used methods employ structural protein data to predict calcium-binding sites. Here we present a method developed to predict calcium-binding proteins and identify calcium-binding sites from protein sequence data using machine learning techniques such as neural networks and support vector machines. We have developed the novel application CalPred, having nine implemented algorithms divided into two filters. The first filter predicts proteins as a whole i.e. whether they are calcium-binding proteins or not, and the second filter predicts the specific calcium-binding sites in the proteins which have passed through the first filter. The tool was able to pick sequences that were calcium-binding in nature but were not picked up by pattern sited as calcium-binding domain in PROSITE database. We also scanned four whole proteomes for potential calcium binding proteins. Addition supplementary information is available at http://www.juit.ac.in/calpred/user_docs.html. Availability: The CalPred tool is available for free use to non-commercial users and can be downloaded to be used in-house as a stand alone server from http://www.juit.ac.in/calpred/index.html