In Silico investigation of anticancer properties of antimicrobial peptides

Objective: This study aims to develop a process using various in-silico methods and tools to predict the potential of antimicrobial peptides (AMPs) in cancer treatment. Through the developed process, the anticancer activity, and physicochemical properties of 10,000 AMPs will be predicted. Additionally, it is anticipated that the results will shed light on the possible dual effects of AMPs. Method: The researcher performed feature selection using the main (training) dataset and trained a machine learning model. This model forms the basis of the study for predicting anticancer activity. The success of the model was measured using fundamental performance metrics. The normal distribution of the results was assessed through the Shapiro-Wilk test, and their performances on different datasets were evaluated using the MANOVA test. Additionally, independent tools, such as clustering of peptide sequences, prediction of physicochemical properties, and three-dimensional homology modelling, were integrated as part of this process. Results: The most successful model was achieved using the random forest algorithm with a combination of amino acid and dipeptide composition features. The developed machine learning model demonstrated an accuracy of 79.5% on the independent (validation) dataset. There was no statistically significant difference in the model's performances between the main and independent datasets. Additionally, out of the 10,000 peptides tested in the experimental dataset, 1648 peptides were discovered to exhibit anticancer activity with optimal physicochemical properties. Conclusion: The findings provide valuable information for a better understanding of the AMPs' anticancer properties and the potential applications of these peptides in cancer treatment.