CPP2Vec: A representation learning approach for cell-penetrating peptides prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Delivery

Background:

• Cell-penetrating peptides (CPPs) are crucial for delivering therapeutic molecules, including antisense oligonucleotides (ASOs), across cell membranes.
• Peptide nucleic acids (PNAs) and phosphorodiamidate morpholino oligomers (PMOs) are key ASOs investigated for Duchenne Muscular Dystrophy (DMD).
• In silico methods are increasingly used to discover novel CPPs, offering a cost-effective alternative to experimental approaches.

Purpose of the Study:

• To develop CPP2Vec, a Word2Vec-based machine learning method for predicting CPPs and their delivery efficiency.
• To create a hybrid dataset (CPP2Vec-GenSet) integrating computational and experimental data to improve CPP representation learning.
• To build and evaluate task-specific models for CPP classification, uptake efficiency prediction, and PMO delivery enhancement.

Main Methods:

• Utilized the Word2Vec technique to learn representations from amino acid sequences of peptides.
• Constructed CPP2Vec-GenSet, a hybrid dataset combining computationally generated and experimentally validated CPPs.
• Developed supervised machine learning models for CPP-Classification, Uptake-Efficiency, and PMO-Delivery, and explored Large Language Models (LLMs) like ProtT5, ProtBERT, and ESM-2 for embeddings.

Main Results:

• CPP2Vec demonstrated robust predictive performance and generalization across classification, efficiency, and delivery tasks.
• The method achieved high computational efficiency compared to existing state-of-the-art CPP prediction tools.
• Alternative models using LLM embeddings (CPP2LLM) were also explored, showing competitive results.

Conclusions:

• CPP2Vec provides a reproducible and efficient in silico tool for identifying and prioritizing CPPs with therapeutic potential.
• The Word2Vec approach effectively learns peptide representations directly from sequence data.
• This ML-based tool supports early-stage research for CPPs relevant to diseases like DMD.