CPPpred-En: Ensemble framework integrating a protein language model and conventional features for highly accurate...

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Drug Discovery and Development

Background:

• Cell-penetrating peptides (CPPs) are crucial for drug delivery due to their cell membrane penetration capabilities.
• Accurate CPP prediction is vital for advancing peptide-based therapies.
• Existing prediction methods often fail to integrate diverse features effectively.

Purpose of the Study:

• To develop an advanced prediction model for cell-penetrating peptides (CPPs).
• To enhance the accuracy and reliability of CPP identification for therapeutic applications.

Main Methods:

• Proposed CPPpred-En, an ensemble learning model integrating conventional peptide features and protein language model (PLM)-based features.
• Evaluated multiple machine learning classifiers and selected optimal feature-classifier combinations.
• Trained and validated the model on the CPP924 and MLCPP 2.0 datasets.

Main Results:

• CPPpred-En achieved high accuracy (97.27% Acc, 0.964 MCC on CPP924; 96.10% Acc, 0.707 MCC on MLCPP 2.0).
• The ensemble strategy demonstrated robust generalization across different datasets.
• Outperformed existing state-of-the-art CPP prediction methods.

Conclusions:

• The integration of conventional and PLM features via ensemble learning is a powerful strategy for improving CPP prediction.
• CPPpred-En serves as a highly accurate and reliable tool for identifying CPPs.
• This advancement holds significant promise for drug delivery and targeted therapy development.