CPPMechPred - Cell-Penetrating Peptide Mechanism Predictor

The first machine learning model for predicting Cell-Penetrating Peptide (CPP) uptake mechanism.

This tool predicts whether a CPP enters cells via endocytosis (energy-dependent) or direct translocation (energy-independent) based on its amino acid sequence.

Try it online: https://huggingface.co/spaces/Numbus00X/CPPMechPred

---

Why does this matter?

Cell-Penetrating Peptides are promising drug delivery vectors, but their therapeutic efficacy depends on how they enter cells:

Mechanism	Process	Implication
Endocytosis	Vesicular uptake	Cargo may be trapped in endosomes
Direct Translocation	Membrane penetration	Direct cytoplasmic access

Comparison with Existing Tools

Tool	CPP Classification	Uptake Efficiency	Uptake Mechanism	Year
MLCPP 2.0	Yes	Yes	No	2022
CPPred-RF	Yes	Yes	No	2017
SkipCPP-Pred	Yes	No	No	2017
CellPPD	Yes	No	No	2013
CPPpred	Yes	No	No	2013
KELM-CPPpred	Yes	No	No	2018
CPPMechPred	No*	No	Yes	2026

*CPPMechPred assumes input peptides are already known CPPs.

No existing tool predicts uptake mechanism. CPPMechPred fills this gap.

---

Quick Start

Installation

git clone https://github.com/Misterbra/cpp-mechanism-predictor.git
cd cpp-mechanism-predictor
pip install -r requirements.txt

Predict a sequence

from src.predict import predict_mechanism

result = predict_mechanism("RRRRRRRR")
print(result)
# {'sequence': 'RRRRRRRR', 'prediction': 'Endocytosis', 'probability': 0.89}

Web Application

streamlit run app.py

Then open http://localhost:8501 in your browser.

---

Results

Model Performance

Validated using nested 5-fold cross-validation with 1000 bootstrap iterations on a non-redundant dataset (80% sequence identity threshold).

Model	AUC-ROC	95% CI	Accuracy	MCC
SVM-RBF	0.795	[0.711 - 0.872]	72.1%	0.447
Logistic Regression	0.786	[0.698 - 0.866]	73.9%	0.474
SVM-Linear	0.753	[0.659 - 0.840]	73.0%	0.456

Comparison with ESM-2 Embeddings

Features	AUC-ROC	# Features
Classical (AAC, physicochemical)	0.834	449
ESM-2 embeddings	0.819	320
Combined	0.830	769

Classical interpretable features perform as well as protein language model embeddings.

Key Predictive Features

Feature	Biological Interpretation
Leucine content	Hydrophobic membrane interaction
Hydrophobicity	Direct translocation preference
Basic residue ratio (R, K)	Endocytosis via HSPG binding
Net charge	Electrostatic membrane interaction

---

Dataset

142 CPPs with experimentally validated uptake mechanisms, curated from peer-reviewed literature.

Class	Count	Examples
Endocytosis	67	TAT, Penetratin, R8, R9
Translocation	75	Melittin, GALA, Magainin 2

After removing sequences with >80% identity: 111 non-redundant peptides.

See supplementary/Table_S1_peptide_references.csv for the complete list with references.

---

Project Structure

cpp-mechanism-predictor/
├── README.md
├── LICENSE
├── requirements.txt
├── app.py                      # Streamlit web application
├── src/
│   ├── features.py             # Feature extraction (449 features)
│   ├── train_mechanism.py      # Model training
│   ├── robust_validation.py    # Nested CV + bootstrap
│   ├── advanced_analysis.py    # t-SNE, error analysis, ESM-2 comparison
│   └── predict.py              # Prediction interface
├── data/
│   └── raw/
│       └── cpp_mechanism_dataset.csv
├── models/
│   └── mechanism_predictor.pkl
├── results/
│   └── [figures and analysis]
└── supplementary/
    └── Table_S1_peptide_references.csv

---

Reproduce Results

# 1. Train the model
python src/train_mechanism.py

# 2. Run robust validation
python src/robust_validation.py

# 3. Run advanced analyses (t-SNE, error analysis, ESM-2)
python src/advanced_analysis.py

---

Citation

If you use this tool in your research, please cite:

@article{brag2026cppmechpred,
  author = {Brag, Nabil},
  title = {Predicting Cell-Penetrating Peptide Uptake Mechanism from Sequence: A Machine Learning Approach},
  year = {2026},
  journal = {Research Square (preprint)},
  url = {https://github.com/Misterbra/cpp-mechanism-predictor}
}

---

License

MIT License - see LICENSE for details.

---

Contact

Author: Nabil Brag

For questions or collaborations, please open an issue on this repository.