A Novel Ensemble Classifier for Protein Secondary Structure Prediction

Understanding the function of proteins requires predicting their secondary structure. There have been improvements over the years, but even now the best predictors can only get around 80% Q8 accuracy using simply the sequence (profile) data. The suggested work deployed a DCNN and SVM ensemble technique to classify subsets of the CullPDB datasets. The protein's secondary structure is hierarchical, with the ability to consider dependencies between neighbouring labels. While experimenting with a newly filtered dataset of proteins from the CullPDB6133 dataset, high accuracy was achieved, with scores of 85% for Q8 Accuracy while using only 80% of the protein residues. The proposed method demonstrates how a Deep Convolutional Neural Network (DCNN) may be used as a generic framework to predict secondary protein structures even when working with few computational resources.