BBPpredict

BBPpredict predicts blood-brain barrier penetrating peptides from peptide sequence information to identify candidates for central nervous system (CNS) drug delivery.

Key Features:

• Training dataset: Uses 326 known BBPs collected from databases and literature and an equal number of non-BBPs sourced from UniProt for model training.
• Independent testing dataset: Validated on an independent test set containing 99 BBPs and 99 non-BBPs.
• Input data: Operates on peptide sequence information as the basis for classification.
• Machine learning comparison: Multiple machine learning methods were compared using nested cross-validation on the training dataset.
• Algorithm: A random forest (RF) algorithm was selected as the final classifier due to superior performance.
• Benchmarking: Demonstrated improvements over previous BBP prediction tools on both training and independent testing datasets.

Scientific Applications:

• BBP identification: Predicts peptides with potential to penetrate the BBB for selection as CNS delivery vectors or therapeutics.
• Discovery acceleration: Accelerates discovery of novel BBPs and complements experimental methods in neuropharmacology and drug discovery.
• Support for CNS drug development: Aids identification of potential drug candidates for CNS diseases.

Methodology:

Multiple machine learning methods were evaluated by nested cross-validation on a training dataset of 326 BBPs and 326 non-BBPs (non-BBPs from UniProt); a random forest model was selected and validated on an independent dataset of 99 BBPs and 99 non-BBPs.