pLM4CPPs: Protein Language Model-Based Predictor for Cell Penetrating Peptides.

Area of Science:

• Computational Biology
• Bioinformatics
• Peptide Science

Background:

• Cell-penetrating peptides (CPPs) are crucial for drug delivery and intracellular targeting due to their membrane permeability.
• Accurate prediction of CPPs is essential to reduce experimental costs and accelerate validation.
• Existing prediction methods require improvement for enhanced reliability and performance.

Purpose of the Study:

• To evaluate the efficacy of various pretrained protein language models (pLMs) in representing CPPs.
• To develop a novel, high-performance deep learning model for CPP classification.
• To establish a reliable computational tool for identifying potential CPPs.

Main Methods:

• Evaluated peptide embeddings from diverse pLMs including ESM variants, ProtT5, and ProtBERT.
• Developed pLM4CCPs, a deep learning architecture employing convolutional neural networks (CNNs) for binary CPP classification.
• Integrated predictions from multiple models to create a consensus classification for improved reliability.

Main Results:

• pLM4CCPs outperformed existing state-of-the-art CPP prediction models in accuracy, MCC, and sensitivity.
• ESM-1280 and ProtT5-XL BFD showed top performance among individual pLMs.
• The pLM4CCPs model demonstrated significant improvements: 4.9-5.5% in accuracy, 9.3-10.2% in MCC, and 14.1-19.6% in sensitivity.

Conclusions:

• The developed pLM4CCPs model offers superior performance for CPP prediction.
• Leveraging consensus predictions from multiple pLMs enhances classification reliability.
• Accessible web server and source code facilitate broader research application in peptide functionality modeling.