Identifying blood-brain barrier peptides by using amino acids physicochemical properties and features fusion method

Blood-brain barrier peptides (BBPs) play a promising role in current drug study of central nervous system diseases. Hence, it is an urgent need to rapidly and accurately discriminating BBPs from non-BBPs. Experimental approaches are the first choice, however, these methods are expensive and take a lot of time. Thus, more and more researchers focused their attention on computational models. In current work, we developed a support vector machine (SVM) based model to identify BBPs. First, amino acids physicochemical properties were employed to represent peptide sequences, and Pearson's correlation coefficient (PCC) and maximal information coefficient (MIC) were applied to extract useful information. Then, similarity network fusion algorithm was utilized to integrate these two different kinds of features. Next, Fisher algorithm was used to pick out the discriminative features. Finally, these selected features were input into SVM for distinguishing BBPs from non-BBPs. The proposed model achieved 100.00% and 89.47% classification accuracies on training and independent datasets, respectively. Additionally, we found that pK2 (NH3) property of amino acid plays a key role in discriminating BBPs from non-BBPs. The results showed that our proposed method is effective, and achieved a significantly improvement in identifying BBPs, as compared with the state-of-the-art approach. The Matlab codes and datasets are freely available at .