Disease-Gene-Prioritization-with-Privileged-Information-and-Heteroscedastic-Dropout

About

This directory contains the code and resources of the following paper:

"Disease Gene Prediction with Privileged Information and Heteroscedastic Dropout". Under review.

• LUPI_RGCN is a relational GNN-based classification algorithm for graphs. It takes disease gene network as inputs and predict the possible associations.
• Our experimental results are based on gene disease network data developed in [1]. The data can be obtained here data

Overview of the Model

We introduce LUPI_RGCN algorithm to address the gene disease prioritization problem. To achieve this goal, we develop a Heteroscedastic Gaussian Dropout algorithm, where the dropout probability of the GNN model is determined by another GNN model with a mirrored GNN architecture. The model is trained under a VAE framework with reparameterization trick.

Sub-directories

• [figures] contains the figures that are used in the paper.
• [LUPI_RGCN] contains implementation of our model for the disease gene network.

Data

There are three datasets in the link.

genes_phenes.mat

• GeneGene_Hs: The HumanNet gene interaction network of size 12331 x 12331.
• GenePhene: a cell array containing Gene-Phenotype networks of 9 species.
• GP_SPECIES: The names of the species corresponding to the networks in 'GenePhene' variable.
• geneIds: The entrezdb ids of genes, corresponding to the rows of the matrix 'GeneGene_Hs' (or 'GenePhene' matrices).
• pheneIds: a cell array containing OMIM ids for phenotypes of 9 species.
• PhenotypeSimilaritiesLog: Similarity network between OMIM diseases.

GeneFeatures.mat

• Microarray expression data - vector of real-valued features for a gene per row (Refer paper for details).

clinicalfeatures_tfidf.mat

• OMIM word-count data - term-document matrix for OMIM diseases (Refer paper for details).

Code Usage

• To run the code, you need python (3.7 I used) installed and other packages, such as pytorch(1.5.0), pytorch-geometric(1.6.1), numpy, pandas, matplotlib.
• The data is .mat form and in order to run it in python, you need to process the data (processing.py) and then run python LUPI_RGCN.py.

License

[1] Li,Y. et al. (2019) Pgcn: Disease gene prioritization by disease and gene embedding through graph convolutional neural networks. bioRxiv.