Tissue-Specific Anticancer Peptide Prediction Using Machine Learning and Pfeature Descriptors

Cancer remains one of the leading causes of mortality worldwide, with the World Health Organization (WHO) reporting approximately 10 million cancer-related deaths in 2020. Cancer is a collective term for a group of diseases characterized by the uncontrolled proliferation of abnormal cells, which can invade or spread to other parts of the body. Traditional treatment approaches, such as radiotherapy and chemotherapy, are widely used but come with significant drawbacks, including severe side effects and nonselective toxicity that affects both healthy and cancerous cells. An alternative and promising therapeutic strategy involves the use of anticancer peptides (ACPs) — short peptides that selectively target and kill cancer cells with minimal toxicity to normal cells. However, identifying novel ACPs through experimental (wet-lab) approaches is both expensive and time-consuming. To address this challenge, this study uniquely focuses on the development of separate machine learning models for prediction of ACPs for six tissue-specific cancer types i.e., breast, cervix, colon, lung, prostate and skin — which offers a more targeted framework. The dataset was sourced from CancerPPD, a database containing experimentally validated ACPs. Our methodology integrates the Pfeature library to compute four composition-based feature descriptors. To select the machine learning algorithms with high performance, we employed LazyPredict which is a rapid and unbiased model screening tool, that allows to select highly promising and common ML algorithms such as Support Vector Machine (SVM) and Random Forest (RF), followed by rigorous hyperparameter tunning. The performance of the models was evaluated using Accuracy, Matthews Correlation Coefficient (MCC) and [Formula: see text]1-score. Additionally, key features associated with high anticancer activity were identified. This study aims to contribute to the field of computational peptide-based drug discovery by providing an efficient and cost-effective method for ACP prediction, ultimately facilitating the development of novel peptide-based targeted cancer therapies.