Anti-parallel Coiled Coils Structure Prediction by Support Vector Machine Classification

Coiled coils is an important 3-D protein structure with two or more stranded alpha-helical motif wounded around to form a “knobs-into-holes” structure. In this paper we propose an SVM classification approach to predict the antiparallel coiled coils structure based on the primary amino acid sequence. The training dataset for the machine learning are collected from SOCKET database which is a SOCKET algorithm predicted coiled coils database. Total 41 sequences of at least two heptad repeats of the anti-parallel coiled coils motif are extracted from 12 proteins as the positive datasets. Total 37 of non coiled coils sequences and parallel coiled coils motif are extracted from 5 proteins as negative datasets. The normalized positional weight matrix on each heptad register a, b, c, d, e, f and g is from SOCKET database and is used to generate the positional weight on each entry. We performed SVM classification using the cross-validated datasets as training and testing groups. Our result shows 73% accuracy on the prediction of anti-parallel coiled coils based on the cross-validated data. The result suggests a useful approach of using SVM to classify the anti-parallel coiled coils based on the primary amino acid sequence.