This repository provides the official implementation of the MOLUNGN model described in our recent study. MOLUNGN is a graph-attention-based deep learning framework developed to integrate multi-omics data (mRNA expression, miRNA expression, DNA methylation profiles) for accurate lung cancer staging classification and stage-specific biomarker identification.

Specifically, MOLUNGN integrates omics-specific Graph Attention Networks (OSGAT) and a Multi-Omics View Correlation Discovery Network (MOVCDN), providing a powerful computational approach to identify critical biomarkers associated with lung cancer progression and facilitating comprehensive systems biology analyses.

To replicate the MOLUNGN computational environment, please use the provided environment.yml file, which contains all necessary dependencies and configurations.

git clone https://github.com/labnjucm/MOLUNGN.git
cd MOLUNGN 

We highly recommend using Anaconda for environment management:

conda env create -f environment.yml
conda activate molungn

Ensure all dependencies are correctly installed by running:

conda list

Our study utilized publicly available multi-omics data from LUAD and LUSC cohorts. You can obtain these datasets from:

The Cancer Genome Atlas (TCGA): https://portal.gdc.cancer.gov

Please follow these steps to preprocess the data:

Download original omics datasets (mRNA expression, DNA methylation, miRNA expression).

Preprocess data into normalized feature matrices according to the procedures described in our paper.

Place processed data into the data/ directory following this structure:

data/
├── LUAD/
│   ├── mRNA.csv
│   ├── miRNA.csv
│   └── DNA_methylation.csv
├── LUSC/
│   ├── mRNA.csv
│   ├── miRNA.csv
│   └── DNA_methylation.csv
└── clinical_labels.csv

Ensure your data follows exactly the formats provided in the sample files in this repository.

Running MOLUNGN After data preparation, here is an example to run the MOLUNGN model with the following command:

python labels_divide.py 

Outputs and Results Results including evaluation metrics, identified biomarkers, and model performance figures will be stored in the results/ directory automatically upon model completion:

Please refer to these results for biomarker information and quantitative evaluation metrics detailed in our published paper.

Repository Structure

MOLUNGN/
├── data_LUAD
├── data_LUSC
├── results/
├── environment.yml
├── labels_divide.py
└── README.md

For questions, concerns, or suggestions regarding MOLUNGN, please contact:

Corresponding Author: [Daifeng Zhang] ([[email protected]])

Lab homepage: https://github.com/labnjucm

License This project is licensed under the MIT License. See the LICENSE file for details.

We hope MOLUNGN provides valuable insights and supports your research on lung cancer classification and biomarker discovery!