Antimicrobial Peptides as Broad-Spectrum Therapeutics: Machine Learning-Based Modelling of Multi-Target Activity

Antimicrobial peptides (AMPs) hold significant potential as broad-spectrum therapeutics because they target various pathogens. Our research objective was to identify chemical, physical, and structural properties that define peptides’ activity against both cancers and viruses. We employed machine-learning techniques, including a support vector machine (SVM) for classification and the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection, to distinguish peptides with single-target activity from those with activity against multiple pathogens. We discovered that specific physicochemical properties were correlated with specific anti-disease activities: anti-virus, anti-cancer, and dual action. Our analysis revealed that specific pairs of amino acids possess the most important specific properties for anti-cancer and anti-virus activity. By developing and automating the hyperparameter tuning process, we optimized feature selection and identified key physicochemical properties associated with broad-spectrum activity. A promising framework for accelerated development of drugs with multiple targets is proposed to benefit experimentalists by pre-selecting specific groups of amino acids based on their targeted reactivity against pathogens. These findings provide valuable insights for designing effective multi-target AMPs as effective cancer drugs.