PHYS 449 Final Project - Systems biology informed deep learning for inferring parameters and hidden dynamics

Group 7 : Richard (Zhifei) Dong, Callum Follett, Nolan Anthony Paul Johnston, Steph Swanson, Jack Zhao

Origninal paper from Yazdani A, Lu L, Raissi M, Karniadakis GE. Systems biology informed deep learning for inferring parameters and hidden dynamics. PLoS computational biology. 2020;16(11):e1007575.

Introduction

This repo contains the code and data to recreate the work by Yazdani et al., 2020. In their work, Yazdani et al. utilized deep learning to determie the parameters for systems biology models of apoptosis, glycolysis, and insulin glucose. In this project, instead of recreating all three models, only two - the apoptosis and glycolysis models - will be studied and tested.

File structure

Three models are tested and validated with the neural network architecture, each with a main.py that is used to train the model and utilizes plot.py to generate plots from the results. There are also two source files nn_gen.py and data_gen.py that contains the neutral network architecture and the code for generating the data for training in the ./src/ directory. Lastly, a param.json can also be found in the ./src/ direcotry that contains all the necessary hyperparameters and may be tuned as needed.

The file structre is outlined below:

• ~/apoptosis
  • main.py
  • plot.py
  • ./src/
    • data_gen.py
    • nn_gen.py
    • param.json
• ~/glycolysis
  • main.py
  • plot.py
  • ./src/
    • data_gen.py
    • nn_gen.py
    • param.json

Dependencies

• sys, argparse, os, datetime
• scipy
• math
• numpy
• torch
• matplotlib

Running each model

To run each model, go to the directory of the model, and use:

python3 main.py

and the model will save the true and measured concentrations for each chemical species, initial loss, all losses, and the predicted p values over epochs as .txt files to ./data

To generate the plots from the .txt files after training, go to the directory of the model, and use:

python3 plot.py

Then the graphs will be save as .pdf files to ./plots/.

Results

All plots generated by this project can be found at https://drive.google.com/drive/folders/1gmoppkCYr93fPJnTrEGddQaoDDXIUjeb