Integrating multiple sequence features for identifying anticancer peptides.

As one of the most terrible diseases, cancer causes millions of deaths worldwide every year. The popular treatment approaches, such as radiotherapy and chemotherapy, have been used in against cancer cells. However, those traditional therapies have side effects on normal cells, time-consuming and expensive. Recent studies showed that anticancer peptides (ACP) may be a potential choice instead of traditional approaches for treating cancer. Therefore, it is desired to develop a computational method to identify anticancer peptides. In this study, a support vector machine (SVM) based computational model was proposed to discriminate anticancer peptides from non-anticancer peptides. In the model, peptide sequences were firstly encoded by amino acids physicochemical (PC) properties and residue pairwise energy content matrix (RECM). Then, Pearson's correlation coefficient, high-order correlation information, and discrete wavelet transform were employed to extract useful information from PC and RECM matrix. The least absolute shrinkage and selection operator (LASSO) algorithm was applied to select discriminative features. Finally, these selected features were fed into SVM for distinguishing ACP from non-ACP. Experimental results demonstrated that the proposed method is powerful, it indicates that our proposed method may be a hopeful tool in discriminating anticancer peptides from non-anticancer peptides. The codes and datasets used in current work are available at https://figshare.com/articles/online_resource/iACP/16866232.