Antiprotozoal peptide prediction using machine learning with effective feature selection techniques.

Area of Science:

• Computational biology
• Drug discovery
• Machine learning in medicine

Background:

• Protozoal infections cause significant mortality and drug resistance, necessitating new therapeutic strategies.
• Antimicrobial peptides are promising drug candidates, but research on antiprotozoal peptides is limited.
• This study addresses the need for predictive tools for antiprotozoal peptides.

Purpose of the Study:

• To develop and validate a machine learning framework for predicting antiprotozoal peptides.
• To classify potential antiprotozoal peptides against diverse negative datasets.

Main Methods:

• Collected experimentally validated antiprotozoal peptides as the positive dataset.
• Utilized multiple negative datasets (non-antimicrobial, antiviral, antibacterial, antifungal, non-protozoal antimicrobial).
• Extracted peptide features using pfeature and selected relevant features with SVC-L1 and mRMR, then applied five classifiers (Decision Tree, Random Forest, SVM, Logistic Regression, XGBoost).

Main Results:

• The XGBoost classifier, using mRMR feature selection, achieved the highest accuracy in predicting antiprotozoal peptides.
• Accuracies ranged from 86.36% (vs. antibacterial) to 97.27% (vs. non-antimicrobial) on the validation dataset.
• The framework effectively differentiates antiprotozoal peptides from various other peptide classes.

Conclusions:

• A robust machine learning framework for predicting antiprotozoal peptides has been successfully developed.
• The developed models are integrated into a user-friendly web server for public access.
• This tool can accelerate the discovery of novel antiprotozoal peptide therapeutics.