CPMP: Cyclic Peptide Membrane Permeability Prediction Using Deep Learning Model Based on Molecular Attention Transformer
The CPMP model is a deep learning approach for predicting the membrane permeability of cyclic peptides. Built on the Molecular Attention Transformer (MAT) neural network, CPMP achieves high determination coefficients (R²) of 0.67 for PAMPA, 0.75 for Caco-2, 0.62 for RRCK, and 0.73 for MDCK permeability predictions.
- Python (3.10)
- Pandas (2.2.3)
- PyTorch (2.5.1)
- RDKit (2024.3.2)
- Scikit-learn (1.6.1)
python predict.py --help
usage: predict.py [-h] [--input_file INPUT_FILE] [--result_file RESULT_FILE]
Predict PAMPA, Caco-2, RRCK and MDCK Membrane Permeability
options:
-h, --help show this help message and exit
--input_file INPUT_FILE (.csv file)
--result_file RESULT_FILE (out file)
The input data input.csv you need to prepare is as follows:
smiles
C/C=C/C[C@@H](C)[C@@H](O)[C@H]1C(=O)N[C@@H](CC)C(=O)N(C)CC(=O)N(C)[C@@H](CC(C)C)C(=O)N[C@@H](C(C)C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N[C@@H](C)C(=O)N[C@H](C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N(C)[C@@H](C(C)C)C(=O)N1C
CC(C)C[C@@H]1NC(=O)[C@@H](CC(C)C)NC(=O)[C@@H](CC(C)C)NC(=O)[C@H](Cc2ccc(O)cc2)NC(=O)[C@@H]2CCCN2C(=O)[C@@H](CC(C)C)NC1=O
You want to save the results to: predict.csv
Run:
python predict.py --input_file input.csv --result_file predict.csv
The obtained results are:
smiles,pampa,caco2,rrck,mdck
C/C=C/C[C@@H](C)[C@@H](O)[C@H]1C(=O)N[C@@H](CC)C(=O)N(C)CC(=O)N(C)[C@@H](CC(C)C)C(=O)N[C@@H](C(C)C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N[C@@H](C)C(=O)N[C@H](C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N(C)[C@@H](CC(C)C)C(=O)N(C)[C@@H](C(C)C)C(=O)N1C,-5.85759162902832,-5.69475793838501,-5.773202419281006,-6.25748348236084
CC(C)C[C@@H]1NC(=O)[C@@H](CC(C)C)NC(=O)[C@@H](CC(C)C)NC(=O)[C@H](Cc2ccc(O)cc2)NC(=O)[C@@H]2CCCN2C(=O)[C@@H](CC(C)C)NC1=O,-6.402388572692871,-6.128831386566162,-5.797394752502441,-5.736731052398682
Before running CPMP on the same data used in the paper, the original data needs to be processed.
For example, if you want to preprocess the PAMPA data, run:
cd data/pampa_uff_ig_true
python process_data.py
This takes around five hour on a regular computer.
Alternatively, pre-processed data (~6.2GB) can be found here. You can directly download and replace the data directory.
For example, if you want to train CPMP model with PAMPA data, run:
python train_pampa.py
For example, if you want to predict PAMPA test data, run:
python predict_pampa.py
The implementation scripts for the baseline methods can be found at this path: baselines/
The CPMP model is built based on the MAT framework. You can find more information about MAT here.
CPMP is released under an Apache v2.0 license.