A deep aggregated model for protein secondary structure prediction. (29th June 2019)

• Record Type: Journal Article
• Authors: Hu, Yu; Nie, Tiezheng; Shen, Derong; Yu, Ge
• Is Part Of: International journal of data mining and bioinformatics. Volume 22:Number 3(2019)
• Page Start: 231
• Page End: 249
• Publication Date: 2019-06-29
• Subjects: sequence translating -- PSSP -- protein secondary structure predication -- machine learning -- cascade model -- deep learning; Data mining -- Periodicals; Bioinformatics -- Periodicals; 006.312
• Languages: English
• ISSNs: 1748-5673

Abstract

Protein sequence analysis is an important research subject that has drawn increasing attention from biomedical researchers. In this research field, Protein Secondary Structure Predication (PSSP) is a significant sub-project for studying protein spatial structure and biochemical function. However, when only the amino acid residues sequence information can be used as the input, it is a challenge problem to predict the spatial structure of the protein. Recently, the deep learning technology achieves great success in information mining. In this paper, we propose a Deep Neural Block Cascade Network (DeepNBCN) for Protein Secondary Structure Predication. This model is constructed by stacking multiple free-adjusted blocks, each for aggregating Feature Extractor Module and Concate and Activate (C&A) Module. The homogeneous and multi-branch architecture can model the complex internal relationship between amino acid sequence and protein secondary structure sequence. We use two publicly available protein datasets to evaluate the proposed model. Experimental results show that our model can obtain 85% Q3 accuracy, 86% SOV score, and 75% Q8 accuracy, respectively, achieving better performance compared with the currently popular predictors.