PractiCPP: a deep learning approach tailored for extremely imbalanced datasets in cell-penetrating peptide prediction

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Cell-penetrating peptides (CPPs) are crucial for effective drug delivery, enabling efficient cellular uptake.
• Identifying CPPs is challenging due to laborious traditional methods and limitations of current computational models with imbalanced datasets.

Purpose of the Study:

• To develop a novel deep-learning framework, PractiCPP, for accurate CPP prediction in highly imbalanced data scenarios.
• To address the limitations of existing models that are not optimized for real-world sparse positive CPP instances.

Main Methods:

• PractiCPP integrates hard negative sampling with advanced feature extraction and prediction modules.
• The framework is specifically designed to learn effectively from imbalanced datasets, including those with extreme positive-to-negative ratios (e.g., 1:1000).

Main Results:

• PractiCPP demonstrates superior performance compared to state-of-the-art methods in CPP prediction.
• Computational validation confirms PractiCPP's high accuracy even with extreme data imbalance.
• Embedding visualizations show that balanced dataset training is inadequate for practical CPP identification.

Conclusions:

• PractiCPP offers a robust solution for CPP prediction, particularly in scenarios with limited positive examples.
• The framework's design accounts for real-world data constraints, potentially accelerating drug delivery research.
• PractiCPP provides a valuable tool for advancing drug delivery system development.