BBPpred

BBPpred predicts blood-brain barrier peptides (BBPs) using a logistic regression classifier and feature representation learning to identify peptides capable of traversing the blood–brain barrier.

Key Features:

• Logistic Regression Classifier: BBPpred employs a logistic regression classifier as its core predictive model.
• Feature Representation Learning: The method utilizes a feature representation learning scheme to extract informative features from amino acid sequence data.
• Feature Investigation and Selection: A comprehensive range of features derived from amino acid sequences is investigated and distilled into informative representations.
• Redundancy Elimination and Feature Reduction: Redundant and irrelevant features are eliminated to select seven key informative features for prediction.
• Benchmark Datasets: Model development and evaluation use two benchmark datasets (one training set and one independent test set) comprising a total of 119 BBPs compiled from public databases and literature.
• Performance Metrics: On the training dataset, BBPpred achieved an AUC of 0.8764 and an AUPR of 0.8757 via 10-fold cross-validation and produced favorable results on the independent test set.

Scientific Applications:

• BBP identification and annotation: Computational screening and annotation of peptides with potential to cross the blood–brain barrier.
• Prioritization for CNS-targeted drug delivery: Prioritizing peptide candidates for development of peptide-mediated delivery to the central nervous system.
• Support for peptide-based therapeutic development: Complementing experimental approaches in characterization and selection of BBPs for neurological therapeutics.

Methodology:

Features derived from amino acid sequences were investigated and processed via a feature representation learning scheme, followed by redundancy elimination to select seven informative features; a logistic regression classifier was trained and evaluated by 10-fold cross-validation on a training set and tested on an independent benchmark dataset consisting of 119 BBPs compiled from public databases and literature.