CycPeptMP

About

Python implementation of CycPeptMP.
CycPeptMP is an accurate and efficient method for predicting the membrane permeability of cyclic peptides.
We designed features for cyclic peptides at the atom, monomer, and peptide levels to concurrently capture both the local sequence variations and global conformational changes in cyclic peptides. We also applied data augmentation techniques at three scales to enhance model training efficiency.

The original cyclic peptide structure (SMILES) and experimentally determined membrane permeability (LogPexp) used in this study were all sourced from CycPeptMPDB.
- Li J., Yanagisawa K., Sugita M., Fujie T., Ohue M., and Akiyama Y. CycPeptMPDB: A Comprehensive Database of Membrane Permeability of Cyclic Peptides, Journal of Chemical Information and Modeling, 63(7): 2240–2250, 2023.
Selected PAMPA datasets used in this research are summarized in all_data.csv.

EXAMPLE.ipynb

Jupyter notebook with an example of prediction.
atoms_model.py

Transformer-based atom model using Node, Bond, Graph, and Conf created from atoms_input.py. The maximum number of heavy atoms in the input is 128.
monomers_model.py

CNN-based monomer model using 16 monomer features created from monomers_input.py. The maximum number of monomers in the input is 16.
peptides_model.py

MLP-based peptide model using 16 peptide features and 2048-bit Morgan fingerprint.

Weights of CycPeptMP (60 times augmentation) for three validation runs (Fusion-60_cv*.cpt).
Weights of fusion model with no augmentation (Fusion-1_cv*.cpt) and 20 times augmentation (Fusion-20_cv*.cpt) for three validation runs in ablation studies.

Li J., Yanagisawa K., and Akiyama Y. CycPeptMP: Enhancing Membrane Permeability Prediction of Cyclic Peptides with Multi-Level Molecular Features and Data Augmentation, Briefings in Bioinformatics, submitted.
Li J., Yanagisawa K., and Akiyama Y. CycPeptMP: Enhancing Membrane Permeability Prediction of Cyclic Peptides with Multi-Level Molecular Features and Data Augmentation, bioRxiv preprint, 2023, 2023.12. 25.573282.