Prediction of amyloid fibrillar aggregates of polypeptide sequences: A soft computing approach

The deposition of amyloid fibrillar aggregates in human brain results in amyloid illnesses. As these aggregates may spread like virus, it is of primary importance to spot such motif regions in protein sequences. Limitations of molecular techniques in identifying them offer sophisticated computational methods for their efficient retrieval. In this paper we tried to enhance the prediction performance of computational approaches by the union of machine learning algorithms: an approach from a soft computing perspective. A filter based dimensionality reduction algorithm has been utilized on the extracted features to obtain a minimal feature subset for Decision tree classification. The filter approach is a multivariate statistical analysis based on the mutual information which is a mixed measure of maximum Relevance and Minimum Redundancy of features. We performed stratified 10-fold cross-validation test to objectively evaluate the accuracy of the predictor.