Protein secondary structure prediction based on fusion of machine learning classifiers

Abstract

Protein secondary structure prediction plays an important role in protein folding and function classification. Although the works available in the literature present good results, protein secondary structure prediction is still an open problem. In this work, we present and discuss a fusion strategy using four different classifiers. The fusion is composed of bidirectional recurrent networks, random forests, Inception-v4 blocks and Inception recurrent networks. In order to evaluate our model, we used CB6133 dataset as training and testing. The fusion achieved 76.4% of Q8 accuracy using the amino acid sequence and similarity information on CB6133, surpassing state-of-the-art approaches.