PYPGLM package is designed to contexualize and model the DBN models corresponding to signaling and gene regulatory systems.
PYPGLM is the new version of the toolbox "FALCON", developed in Matlab in 2017 and described in De Landtsheer et al. in Bioinformatics: here is the link.
Python >=3.12 is required.
- networkx
- numpy
- pandas
- emoji
- pytest
- pytest-cov
- openpyxl
- scipy
- matplotlib
- gaft
- seaborn
- joblib
- scikit-learn
You can install PYPGLM directly from PyPI using pip:
pip install PYPGLM - You can also clone the repo to your local machine, set up a virtual environment, install the requirements from the 'requirements.txt' file.
To correctly model the network with its experimental data:
- The network topology file (.txt, .xls, .xlsx, .csv, and .sif) should contain the interactions of nodes within the investigated network, specifying their type and gate.
- Experimental data (xls, xlsx, csv) should consist of 3 tables representing the measured input and output node values and their corresponding errors listed in columns across all experimental conditions listed successively in rows. Regarding the file in .csv format, the 3 tables should be given as 3 individual files, while in the case of .xls(x) format, the 3 tables should be presented as 3 distinct sheets within the same corresponding file.
Examples demonstrating how to run the optimization and regularization are provided in 'Driver_PYPGLM_Optimization' and 'Driver_PYPGLM_Regularization' files, respectively." here is the link
The optimization algorithms within PYPGLM are 'SLSQP', 'L-BFGS-B', and 'Trust-constr'.
The regularization algorithm within PYPGLM includes:
- L1: simple L1 norm on parameters. This induces a pruning of less important inhibitory edges, but has no effect on stimulatory edges
- L2: Induces a decrease of weights of unimportant inhibitory edges
- L1_groups: Induces an equalization of edge values between contexts when these are discrete (different cell lines)
Developers:
Salma Bayoumi, [email protected] [email protected]
Sebastien De Landtsheer, [email protected] [email protected]