Speeding Up the Virtual Design and Screening of Therapeutic Peptides: Simultaneous Prediction of Anticancer Activity and Cytotoxicity

Abstract In this chapter, we propose a novel computational methodology for the virtual design and screening of peptides with potential anticancer activity against different cancer cell lines, and low cytotoxicity against diverse healthy mammalian cells. In this context, a multitasking (mtk) chemoinformatic model combining Broto–Moreau autocorrelations with artificial neural networks was derived from a data set containing 1933 cases of peptides. The model exhibited an accuracy greater than 92% in both training and prediction (test) sets. A simple statistical approach was applied to qualitatively correlate the changes in the physicochemical properties (molecular descriptors) of the peptides with the corresponding variations in their biological effects. To illustrate the practical use of the proposed in silico methodology, 12 peptides were designed and predicted by the mtk-chemoinformatic model. Encouraging results were obtained, indicating that these peptides can be considered for future experiments focused on the assessment of anticancer activity and cytotoxicity.