Fogli Iseppe et al. 2021 male and female human epicardial ventricular cardiomyocyte models and ML code.

The parent model developed by Yang and Clancy (Yang, P.-C. & Clancy, C. E. In silico Prediction of Sex-Based Differences in Human Susceptibility to Cardiac Ventricular Tachyarrhythmias. Front. Physiol. 3, 1–12 (2012)) was modified by:

• increasing the maximal transport rate of the Na+-Ca2+ exchanger (NCX) by 15% in the female model
• removing the original female-to-male difference in SERCA formulation
• removing the original female-to-male difference in Na+/K+-ATPase formulation The updated model better recapitulates observed functional sex differences in Ca2+ handling.

---

Contents:

• readme.md this file

"AFI_model" folder

• run_AFI_model.py runs the simulations with the baseline male and female models
• AFI_ODEfile.cpp source file containing ODEs for the Fogli Iseppe et al. model
• AFI_ODEfile.out compiled file containing ODEs for the Fogli Iseppe et al. model: icpc -O3 AFI_ODEfile.cpp -o AFI_ODEfile.out. May require #include <sstream> if using g++.
• APInfo.hpp helper file for metrics extraction
• Drug_effects_all_*X.dat parameters for simulating the drug set at different doses

"create_datasets" folder

• create_df_all_with_EAD_flag.py generates the datasets adding the info about EAD presence
• create_training_and_test_dfs.py generates the training and test datasets
• all_drugs_epi_july2020.xlsx file containing data related to the simulated drugs and presence of EADs in the simulations

"RFE" folder

• rfe_lr_bayes_optimized_repeated_n_times.py runs the RFE algorithm with a Logistic Regression model
• rfe_svm_bayes_optimized_repeated_n_times.py runs the RFE algorithm with a Support Vector Machine model

---

Instructions

List of packages required

• Numpy
• Pandas
• Multiprocessing
• Joblib
• Scikit-Learn
• Hyperopt

List of programming environment

• Python (tested with python 3.7.3)
• C++ compiler (tested with intel compiler icpc)

Run the simulations

1. Enter in the AFI_model folder.

2. Create a data folder.

3. Run the python script named run_AFI_model.py.

Create the datasets

1. Make a copy of the all_drugs_epi_july2020.xlsx file in the AFI_model folder.

2. The create_df_all_with_EAD_flag.py script expects a specific organization of the folders containing the data. In detail, the directory tree inside the data folder should follow the structure below:

• Sex folder (Male or Female)
• Hormone folder (NH for no hormones, DHT_10 for low level of testosterone, DHT_35 for high level of testosterone, EF for early follicular phase, LF for late follicular phase, LU for luteal phase, see run_AFI_model.py for hormone details, default is NH).
  Create these new folders in data and move the simulated data in the appropriate sex- and hormone-specific subfolder, e.g.,
• Male/NH/BCL.500.***
• Female/NH/BCL.500.***

3. Create a copy of the create_df_all_with_EAD_flag.py script in each sex- and hormone-specific subfolder and run the script python create_df_all_with_EAD_flag.py . The script will generate a csv file containing the measured biomarkers.

4. Create a copy of the create_training_and_test_dfs.py script in each sex- and hormone-specific subfolder and run the script python create_training_and_test_dfs.py. The script will split the previously created csv file in two files containing the delta biomarkers (drug value - control value) for the training (high risk and safe drugs) and test (intermediate risk) sets.

Run the recursive feature elimination algorithm

1. Enter in the RFE folder.

2. Create two folders under REF

• data folder.
• results folder

3. Move the training csv files created during the previous step in the data folder.

4. Run the python script named rfe_lr_bayes_optimized_repeated_n_times.py if you want to train a logistic regression classifier, or rfe_svm_bayes_optimized_repeated_n_times.py if you want to train a support vector machine classifier. May require pip install lightgbm scikit-learn if lightgbm is not present.

5. Customizable variables: n_jobs is the maximal number of jobs (parallelization) the script will do, the range function on the same line of code (108 for lr, 109 for svm) decides how many times you want to repeat the entire RFE process for each input file, n_iter is the number of iterations for each optimization step (I generally used 200).

---

Reference

Fogli Iseppe A., Ni H., Zhu S., Zhang X., Coppini R., Yang P.C., Srivatsa U., Clancy C.E., Edwards A.G., Morotti S., Grandi E.. Sex‐specific classification of drug‐induced Torsade de Pointes susceptibility using cardiac simulations and machine learning. Clinical Pharmacology & Therapeutics. 2021 Mar 26; TO BE COMPLETED WITH VOLUME AND PAGES ONCE PUBLISHED. doi: https://doi.org/10.1002/cpt.2240

Please, cite the above paper when using these codes.