The protein structure prediction module of the prot-grid information system

Abstract

In this work, we describe the protein secondary structure prediction module of a distributed bio-informatics system. Protein databases contain over a million of sequenced proteins, however there is structuring information for at most 2% of that number. The challenge is to reliably predict the structure based on classifiers. Our contribution is the evaluation of architectures of multiple classifier systems on a standard dataset (CB- 396) containing protein sequencing information. We compare the results of a single classifier system based on SVMs, as well as with our version of an SVM based adaBoost algorithm and a novel fuzzy multi-SVM classifier.