GGOT: Detecting Disease Critical Transitions Using Gaussian Graphical Optimal Transport Embedded With Protein Interaction Networks.

This repository contains our proposed methods Gaussian Graphical Optimal Transport (GGOT) for detecting critical transitions and identifying trigger molecules of diseases, the corresponding pre-processed diseases datasets from TCGA and GEO databases, and pre-processed Protein-Protein Interaction (PPI) Networks of the Human and Mouse Genomes.

Code for the paper:

Wenbo Hua, Ruixia Cui, Heran Yang, Jingyao Zhang, Chang Liu, Jian Sun. "Detecting Disease Critical Transitions Using Gaussian Graphical Optimal Transport Embedded With Protein Interaction Networks"

Overview

GGOT uses Gaussian graphical model, incorporating the gene interaction network, to model the data distributions at different disease stages. Then we use population-level optimal transport to calculate the Wasserstein distance and transport maps between stages, enabling us to detect critical transitions. By analyzing the per-molecule transport map, we quantify the importance of each molecule and identify the trigger molecules. Moreover, GGOT predicts the occurrence of critical transitions in unseen samples and visualizes the disease progression process.

Datasets

Clicking on the name of the corresponding dataset will redirect you to the website to download the corresponding dataset. The dataset of XJTUSepsis needs to contact [email protected] to get access.

Dataset	Description	Species	stages	Type of disease
GSE48452	non-alcoholic fatty liver	Home sapiens	4	Chronic progressive benign disease
GSE2565	lung injury	Mus musculus	10	Acute progressive noncritical disease
LUAD	lung adenocarcinoma	Home sapiens	8	Chronic progressive malignant disease
COAD	colon adenocarcinoma	Home sapiens	8	Chronic progressive malignant disease
XJTUSepsis	sepsis	Home sapiens	8	Acute progressive critical disease
GSE154918	sepsis	Home sapiens	4	Acute progressive critical disease

Prerequisites

1 Create your environment and install required packages

conda env create -f environment.yaml -n yourenv

2 Download the pre-processed Protein-Protein Interaction (PPI) Networks of the Human and Mouse Genomes into "PPI/". We construct the PPI networks using STRING database.

PPI Networks of two species:[GGOT_PPI]

3 Download the RNA-seq dataset of different diseases into "data/source/".

Raw datasets: [GGOT_datasets]

Make sure your data file tree is similar to the following.

GGOT
├──run_model.py
├──data/
│  ├──source
│  │  ├──GSE48452
│  │  │  ├──GSE48452.csv (RNA-seq expression)
│  │  │  ├──group.csv (stage information)
│  │  ├──GSE2565
│  │  │  ├──GSE48452.csv
│  │  │  ├──group.csv
│  │  ├──LUAD
│  │  │  ├──LUAD.csv
│  │  │  ├──group.csv
│  │  ├──......
│  ├──......
├──PPI
│  ├──Human
│  │  ├──ppi_database.npy
│  │  ├──map_gene_protein.csv
│  │  ├──map_gene_protein_full.csv
│  │  ├──symbol2id.csv
│  ├──Mus
│  │  ├──ppi_database.npy
│  │  ├──map_gene_protein.csv
│  │  ├──map_gene_protein_full.csv
│  │  ├──symbol2id.csv
├──......

Running GGOT for datasets with different types

• running model for the chronic progressive non-critical disease, GSE48452

python run_model.py -d GSE48452 -s Human

• running model for the acute progressive non-critical disease, GSE2565

python run_model.py -d GSE2565 -s Mus

• running model for the chronic progressive critical disease, LUAD, COAD

python run_model.py -d LUAD -s Human
python run_model.py -d COAD -s Human

• running model for the acute progressive critical disease, XJTUSepsis, GSE154918

python run_model.py -d XJTUSepsis -s Human
python run_model.py -d GSE154918 -s Human

Making the visualization

python visualization.py -d GSE2565

Results (example on simulation dataset)

If you have any problems with this repository, please send emails to [email protected] for discussions.