python version=3.11.7
conda create -n HGGS python=3.11.7conda activate HGGS
pip install -r requirements.txttrain root dirctory:
src/yu/tasks/OnlineSamplingTools
The Baseline_Training_code folder contains the baseline scripts
Config: training config details
The Online_Sampling folder contains the HGGS scripts for training
(Gradient-based [Ada_Gradient_init.py] and Multigrid Genetic Sampling[Genetic_Sampling.py])
The Sampling_Algorithm folder contains HGGS scripts for sampling
utils directory:
src/yu/tasks/const: Definition of constant StrEnum data, src/yu/tasks/core: Configuration utilities for auto-saving, src/yu/tasks/exception: Custom-defined exceptions, src/yu/tasks/nn: Neural network definitions, src/yu/tasks/tools: Script functions definition.
ODEs directory:
Brusselator system:
src/yu/tasks/model_brusselator.pyCell Cycle system:
src/yu/tasks/model_HSECC.py,MPF system:
src/yu/tasks/model_MPF_2_Var.py,Activator Inhibitor system:
src/yu/tasks/model_PS2_01.py,and the general model
src/yu/tasks/ode_model.py.The oscillatory frequency is calculated in
src/yu/tasks/pde_models.py.
data root dirctory: data/
2param: The Brusselator system (2 system coefficients)
6param: Data for 3 biological systems: Cell Cycle (HSECC), Mitotic Promoting Factor (MPF_2_Var), and Activator Inhibitor (PS2_01)
train/sampling config dirctory: resource/
Training config: OnlineSamplingTools/train.env and src/yu/tasks/BioSysConfig/BioSysConfig.py
sample config (LHS): sample/multi_generate_data.env [For multithreaded LHS]
pretrained model dirctory: output/mlp/reg_2/
xparam/system_name/model_name/seed/best_network.pth
We have provided the training and test datasets (located in the data/ directory). If you wish to skip this step and proceed with training the model, please refer to section 4.
An Example: Cell Cycle System
LHS produce data:
LHS_code/HSECC.env
LHS code
1. modify resource/sample/multi_generate_data.env
2. cd src/yu/tasks/sample/
3. python multi_generate_data.pyCitation
Prigogine, I. (1978). Time, structure, and fluctuations. Science, 201(4358), 777-785. Time, Structure, and Fluctuations | Science
In the code, we call it brusselator
Citation
Liu, Z., Pu, Y., Li, F., Shaffer, C. A., Hoops, S., Tyson, J. J., & Cao, Y. (2012). Hybrid modeling and simulation of stochastic effects on progression through the eukaryotic cell cycle. The Journal of chemical physics, 136(3). Hybrid modeling and simulation of stochastic effects on progression through the eukaryotic cell cycle - PMC (nih.gov)
Called HSECC in code
Citation
Novak, B., & Tyson, J. J. (1993). Modeling the cell division cycle: M-phase trigger, oscillations, and size control. Journal of theoretical biology, 165(1), 101-134. Modeling the Cell Division Cycle: M-phase Trigger, Oscillations, and Size Control - ScienceDirect
Called MPF_2_Var in code
Citation
Murray, J. D. (2007). Mathematical biology: I. An introduction (Vol. 17). Springer Science & Business Media. Mathematical Biology: I. An Introduction | SpringerLink
Called PS2_01 in code
4.1 Cloning this repo
git clone https://github.com/zhaoxixixi/HGGS.git4.2 Getting Started
Create a new virtual enviroment for HGGS.
Navigate to the directory for model training/testing.
cd src/yu/tasks/OnlineSamplingTools4.3 Training
# HGGS-1w
python multi_seed_train.py "HGGS-1w" --model_path "/HSECC/" \
--seeds "[53]" --epoch_n 3000 --warm_up_epoch 200 --last_epoch_n 2000 --base_lr 0.0025 \
--ode_model_name HSECC --total_training_samples 10000 --batch_size 40960This code will save the model in this path from the root code directory: project-root/output/mlp/reg_2/6param/HSECC/HGGS-1w/53
The remaining training scripts are located in (from the root) project-root/src/yu/tasks/train_code
4.4 Testing
testing for RMSE results
python test.py --train_set "HGGS-1w" --model_path "HSECC/" --ode_model_name "HSECC" \
--seeds "[53]"You should provide the pretrained model here (relative to the root code folder): project-root/output/mlp/reg_2/6param/HSECC/HGGS-1w/53/best_network.pth
The testing scripts are located in (relative to the root code folder) project-root/src/yu/tasks/test_code.
testing for Imbalance Ratio and Gini Index results
python diversity.py --dir_path "../../../../output/mlp/reg_2/6param/HSECC" --model_names "HGGS-1w" \
--seeds "[53]" --system_dimension 65.1 download output.zip
All pretrained models are provided at https://drive.google.com/file/d/16bXqwkUnSJ9p--B76aoMqujarpQRsHyM/view?usp=sharing
5.2 unzip
# move output.zip to project-root folder
unzip output.zipExample: pretrained model directory output/mlp/reg_2/
x_param/system_name/model_name/seed/best_network.pth