Transmembrane Protein Prediction Using N-Gram and Random Forests

With recent development of proteomics, the importance of transmembrane proteins has been widely acknowledged. Previous bioinformatics studies have mainly focused on the classification of membrane proteins and ignored the important role of transmembrane proteins. In this study, we integrated the preceding order of amino acids into a machine learning approach based on a revamped N-gram model to predict transmembrane proteins using only protein sequence information. The framework consists of two steps: The N-gram model revamped for processing protein sequences was used as a feature extraction algorithm; then, we compared the performance of the popular classifiers logistic regression, Random Forests, support vector machine, and K-nearest neighbor using the N-gram model. N-gram combined with the Random Forests classifier obtained the highest accuracy at 95.6%, which is higher than other methods. The finding can help future studies on the structure and function of transmembrane proteins, drug design, and the classification of membrane proteins. In addition, a publicly accessible web server and software was established.