Dynamics-based Feature Augmentation of Graph Neural Networks for Variant Emergence Prediction

Setup Environment

To set up the environment for the GNN Variants project, use the provided environment.yml file. Create a Conda environment using the following commands:

conda env create -f environment.yml
conda activate projectGNN

This will create a conda environment named projectGNN with the specified dependencies.

Configuration

All variable configurations are specified in the config.json file. Adjust the values in this file according to your requirements:

• is_graph: Boolean indicating whether the data represents a graph (true) or not (false).

• ADJ_bool: Boolean indicating whether the graphs uses country adjacency to construct the adjacency matrix (true) or not (false).

• Flights_bool: Boolean indicating whether flights data is used to construct the adjacency matrix (true) or not (false).

• self_loops: Boolean indicating whether self-loops are allowed in the graph (true) or not (false).

• EW_bool: Boolean indicating whether edge weights are present in the graph (true) or not (false).

• topX_bool: Boolean indicating whether to filter noisy data by using data from the top X% of countries which reported cases (true) or not (false).

• topX_C: An integer specifying the value of X when topX_bool is true.

• dom_thresh: A floating-point value representing the dominance threshold.

• use_r: Boolean indicating whether to use 'r' (true) or 'p' (false).

• use_S: Boolean indicating whether to use 'S' (true) or not (false).

• min_epochs: Minimum number of training epochs.

• max_epochs: Maximum number of training epochs.

• T: An integer representing the value of timesteps to be used a features.

• reg_bool: Boolean indicating whether we are doing a regression (true) or classification (false).

• early_stopper_patience: An integer specifying the patience for early stopping.

• early_stopper_delta: An integer specifying the delta for early stopping.

• variants_path: Path to the file containing variant data (data/all_vars21_clustered_NEW.csv). This can be edited to evaluate on less variants.

• countries_path: Path to the file containing country data (data/countries_clustered.csv). This can be edited to train/evaluate on less countries.

• device: Device for computation ("cpu" or "gpu"). GPU IS CURRENTLY NOT SUPPORTED

• ITERATION_NAME: A string specifying the name of the iteration ("GNN_Test") Results will be saved in a folder with this name followed by a timestamp.

• IS_DEBUG: Boolean indicating whether debugging mode is enabled (true) or not (false).

Modify the config.json file per your preferences, and run the below command to set the config variables:

python modulefuncs/config.py config.json

Running the Main Script

The main script, main.py, is the entry point for your project. Execute it using the following command:

python main.py

Please make sure you have activated the Conda environment (as described here), and set the config variables (as described here) before running the script.

Custom Models

Define custom models in the models.py file. The model to be used in main.py should be implemented as class ModelM.

Visualization of Results

After training, visualize the results using the viz_script.py file. Provide the directory where the results are stored by modifying the variables directory_name and csv_directory

Other Directories

Data

This contains all the data used for the various tasks. Data is up to date as of December 17th, 2023

Data Preprocessing

The following contains the scripts used to preprocess the data. This includes cleaning the data, getting interpolated data, and calculating S values.

Results

All iterations/runs are saved in this directory.

Experiments

All experiments ran for our paper can be found in scripts here. Inside the baseline_new directory, we see the matlab code for the trivial and baseline regression models. others correspond to:

• DT.py: Decision Tree Classifier
• DT_regression.py: Decision Tree Regression
• GNN_LOSS_0.3.py: DIL-GCN. Edit p variable to select desired regularization.
• MLP.py: MLP Classifier
• MLP_Regression.py: MLP Regression
• RNN2.py: GRU Regression and Classification
• RNN_GNN.py: T-GCN Regression and Classification
• base.py: Trivial and Adjacency Based Classifiers
• embVar_GNNScripy.py: EE-GCN Regression and Classification
• encVar_GNNScripy.py: AE-GCN Regression and Classification

★★★ For any questions, please email Srikar Mutnuri (nmp8rj@virginia.edu) or Mansooreh Montazerin (mmontaze@usc.edu).