CPPred-FL: a sequence-based predictor for large-scale identification of cell-penetrating peptides by feature...

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Delivery Systems

Background:

• Cell-penetrating peptides (CPPs) are crucial for intracellular cargo delivery, with significant therapeutic potential.
• Accurate identification of CPPs is essential for understanding their mechanisms and therapeutic applications.
• Machine learning approaches are increasingly used for CPP identification, but effective feature representation remains a challenge.

Purpose of the Study:

• To develop a powerful bioinformatics tool, CPPred-FL, for fast, accurate, and large-scale identification of CPPs.
• To introduce a novel feature representation learning scheme to improve predictive performance by exploiting CPP characteristics.
• To optimize feature selection for enhanced predictive accuracy and efficiency.

Main Methods:

• Developed CPPred-FL, a tool integrating 45 random forest models with diverse feature descriptors (compositional, positional, physicochemical).
• Implemented a feature representation learning scheme incorporating class and probabilistic information.
• Performed feature space optimization to remove redundant features, utilizing 19 informative features.

Main Results:

• CPPred-FL demonstrated superior performance compared to state-of-the-art predictors in benchmarking experiments.
• The novel feature representation learning scheme effectively captured essential CPP characteristics.
• Feature selection resulted in a highly efficient model using only 19 informative features.

Conclusions:

• CPPred-FL is a highly effective tool for large-scale CPP identification.
• The developed feature representation and selection methods significantly enhance prediction accuracy.
• CPPred-FL will accelerate CPP characterization and their clinical therapeutic applications.