GMHAN

Installation

GMHAN can be downloaded by

git clone git@github.com:mping315/GMHAN.git

Installation has been tested in the Linux platform

Data

All the data we used are stored in the './data/' folder. Among them, the gene data is stored in the ./gene folder, and the miRNA data is stored in the ./mirna folder.

./data/gene/total_feature_500.csv: We extracted features of genes using EMOGI and node2vec. For each of the 16 cancer types, we calculated mutation frequency, DNA methylation difference rate, and gene differential expression value as multi-omics features. We then ran node2vec on a gene-gene network to obtain gene topological features.

./data/gene/gene_graph.bin: This file contains the gene-gene and gene-miRNA networks we built. The gene-gene network was taken from the CPDB database, and the gene-miRNA network was built using data from miRTarBase, TarBase, and miRWalk database.

./data/gene/gene_trainval_fold_0.txt: Sample labels for the gene training set. ./data/gene/gene_trainval_mask_fold_0.txt: Sample index for the gene training set. ./data/gene/gene_test_fold_0.txt: Sample labels for the gene test set. ./data/gene/gene_test_mask_fold_0.txt: Sample index for the gene test set.

./data/miRNA/6.feature_miRNA_index_total.csv: This file contains the miRNA features we extracted. For each of the 16 cancer types, we calculated the mean z-score of miRNA expression levels, the mean of miRNA expression values, and a GIP similarity matrix based on miRNA-disease associations.

./data/miRNA/miRNA_graph.bin: This file contains the miRNA-miRNA and gene-miRNA networks we built. The miRNA-miRNA network was constructed using Pearson similarity between miRNAs, and the gene-miRNA network was taken from miRTarBase, TarBase, and miRWalk.

./data/miRNA/mirna_trainval_fold_0.txt: Sample labels for the miRNA training set. ./data/miRNA/mirna_trainval_mask_fold_0.txt: Sample index for the miRNA training set. ./data/miRNA/mirna_test_fold_0.txt: Sample labels for the miRNA test set. ./data/miRNA/mirna_test_mask_fold_0.txt: Sample index for the miRNA test set.

Overview

Here, we provied an implementation of GMHAN, in Pytorch and Pytorch Geometric.

• propress_deepwalk.py: Gene feature pretreatment.
• model.py: Defines the core architecture of GMHAN.
• utils.py: Functions used in GMHAN.
• train and test.ipynb: Train and evaluate the model.

Requirements

• Python 3.7
• Pytorch 1.12.1
• Pyroch Geometric 2.0.4
• numpy 1.18.5
• pandas 1.3.5