CpACpP

CpACpP predicts the activity of cell-penetrating anticancer peptides (Cp-ACPs) using machine-learning models to prioritize candidate Cp-ACPs for cancer research.

Key Features:

• Machine Learning Algorithms: Uses Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost) trained on experimentally validated datasets to predict Cp-ACPs.
• Dual Subpredictors: Implements two independent subpredictors specifically for cell-penetrating peptides (CPP) and anticancer peptides (ACP) to capture distinct peptide characteristics.
• Feature Selection via RFE: Applies a multilayered recursive feature elimination (RFE) approach to select and combine compositional and physicochemical features from CPP and ACP datasets.
• Performance Metrics and Benchmarking: Reports superior accuracy compared to ACPred, mACPpred, MLCPP, and CPPred-RF, with ACP subpredictors achieving ACC of 0.98 and AUC of approximately 0.98.
• CPP Predictor Performance: Reports CPP predictor values around 0.94 (hybrid-based features) and 0.95 (independent datasets).
• Independent Sequence Evaluation: Achieves accuracies of about 0.79 for CPP classifiers and 0.89 for ACP classifiers on independent sequences.
• Consensus-Based Fusion: Employs a consensus-based fusion method that attains an AUC of 0.94 for predicting Cp-ACP activity.

Scientific Applications:

• Cp-ACP identification: Prioritizes candidate cell-penetrating anticancer peptides from sequence datasets for downstream experimental validation.
• Peptide design and discovery: Guides the design and selection of novel Cp-ACPs to accelerate discovery of therapeutic peptide candidates against solid tumors and hematologic malignancies.

Methodology:

Train RF, SVM, and XGBoost models on experimentally validated CPP and ACP datasets, apply multilayered RFE for feature selection of compositional and physicochemical attributes, and combine predictions via a consensus-based fusion method.