Machine learning-aided screening framework for wound healing peptides

Abstract

Chronic wounds characterized by prolonged inflammation and persistent infection pose a significant burden to global healthcare systems. Currently, antibiotics and non-steroidal anti-inflammatory drugs (NSAIDs) are administered to patients. Prolonged use of antibiotics is severely discouraged owing to the rapid rise in antimicrobial resistance, and the use of NSAIDs can also increase the risk of infection. Thus, the discovery of novel therapeutics for chronic wounds is crucial. Antimicrobial peptides (AMPs) are an emerging class of therapeutics, which are effective and has no known mechanism of inducing resistance. Anti-inflammatory peptides (AIPs) are another class of therapeutic peptides that can reduce inflammation by eliciting anti-inflammatory cytokine response. A single peptide possessing both AMP and AIP activities can be an ideal therapeutic for the treatment of chronic wounds. However, the discovery of peptides with multiple properties via experimental testing is a daunting task. In this work, we propose a classification framework using machine learning for the identification of wound healing peptides (WHPs) i.e., which possess both AMP and AIP activities. The proposed framework uses XGBoost algorithm with amino-acid composition, sequence analysis and physicochemical properties as feature representation methods (FRMs) to develop binary classifiers. The model developed by combining all the three FRMs, resulted in the highest accuracy of 93.3 and 76.2% for AMP and AIP classifications, respectively. An easy-to-use freely accessible web tool (WHP-Pred) has also been developed.