PredAPP

PredAPP predicts antiparasitic peptides to enable large-scale identification of antiparasitic candidates.

Key Features:

• Machine Learning-Based Prediction: Employs an ensemble of machine learning (ML) classifiers to predict antiparasitic peptides.
• Balanced Training Dataset: Uses an undersampling method based on cluster centroids to address class imbalance in the training data.
• Feature Engineering and Selection: Integrates nine distinct groups of features with six ML algorithms to generate 54 classifiers and selects the best-performing feature representation within each group.
• Model Integration: Integrates selected feature representations using logistic regression to construct the final predictive model.

Scientific Applications:

• Large-scale antiparasitic peptide screening: Supports genome- or peptide-library-scale identification of candidate antiparasitic peptides.
• Benchmarking and comparative assessment: Reports accuracy of 0.880 and AUC of 0.922 and is compared against AMPfun (accuracy 0.739, AUC 0.873).

Methodology:

Generation of a balanced training dataset via centroid-based undersampling; combination of nine feature groups with six ML algorithms to produce 54 classifiers; selection of optimal feature representations based on performance metrics; integration of selected representations through logistic regression to form the final model.