A New Direction in Membranolytic Anticancer Peptides classification: Combining Spaced k-mers with Chaos Game Representation.

Abstract

The role of Membranolytic Anticancer Peptides (ACPs) as breast cancer therapeutics is gaining popularity because of their capacity to delay cellular resistance development and eliminate some common chemotherapy challenges, like cytotoxicity, aftereffect, etc. As breast cancer is a leading cause of death among women globally and designing effective treatment mechanisms for it is a crucial step, identifying the potent ACPs is a vital contribution towards cancer treatment. Since ACPs are small protein sequences, it is essential to devise efficient and effective sequence analysis methods for them. In this regard, Chaos game representation (CGR) is a popular technique used to visualize protein sequences as 2D images. In CGR, amino acids of a sequence are projected on a pre-defined axis, and the resulting image is used as input for machine learning/deep learning models. The traditional method of projecting amino acids uses the sine and cosine functions, but in this study, we explore the use of secant and cosecant functions as an alternative. Moreover, we use Spaced k-mers of a sequence as an alternate way to manipulate its amino acids rather than the original sequence, which is proven to show better results than traditional methods which deal with each amino acid within the sequence separately. In this study, we propose a new approach for ACP classification based on the concept of CGR and spaced k-mers. We generate images of the peptide sequences and use them as input to the deep learning (DL) models to perform classification. The results show that our proposed approach achieves high predictive performance for ACP classification, and provides a new direction for identifying the potent ACPs which can be used for breast cancer treatment.