Membrane Protein Transmembrane Helix Prediction, Based On Feature Fusion

Since 1990s, the Human Genome Project has greatly accelerated the development of genomics and proteomics, and people have obtained vast amounts of data of amino acid sequences and protein sequences. However, the protein structure determination can not keep up the pace of the rapid growing of protein sequence data. Membrane proteins account for about 30% of the entire protein sequences, but only 1% of them whose structures have been determined. As part of the cell membrane, the membrane protein has important functionalities such as communication between cells, regulating solute and ion transport inside and outside the cell membrane, the cell's "sense organs" and other functions, which plays an important role in maintaining the normal functioning of the organism. Because of a close relationship between protein structure and function, if the structure can be determined, we can obtain its function. In addition, membrane protein receptors are the drug targets of most medicines, thus, if we can obtain the structural features of membrane protein receptors, there will be huge benefits in new drug development.This paper focuses on a subtask of protein structure prediction-transmembrane helices prediction. Previous studies have shown that the position-specific scoring matrix (PSSM) based features are suite for protein attributes prediction. However, PSSM only describes the sequence evolutionary information and there are many other physical and chemical characteristics of the amino acids in membrane proteins which can potentially further improve the prediction accuracy. In this study, we perform transmembrane helices prediction based on feature fusion by effectively integrating the PSSM, hydrophobicity, hydrophilicity, structure tendency and distribution preference features.Experimental results on benchmark dataset show that the proposed method improves all the five widely used evaluation indexes for transmembrane helices prediction, thus demonstrate the proposed method is suite for transmembrane helices prediction.