HetIG-PreDiG: A Heterogeneous Integrated Graph Model for Predicting Human Disease Genes based on gene expression

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Graph analytical methods are used to identify novel genes in complex diseases.
• Existing methods face limitations including ignoring unconnected nodes, using homogeneous graphs, relying on incomplete data, and inadequate evaluation of predictions.

Purpose of the Study:

• To address the limitations of current graph analytical approaches for disease gene identification.
• To develop a robust model for predicting novel disease genes with high accuracy.

Main Methods:

• Developed the Heterogeneous Integrated Graph for Predicting Disease Genes (HetIG-PreDiG) model incorporating gene-gene, gene-disease, and gene-tissue associations.
• Utilized low-dimensional node representation and a Gene-Disease Prioritization Score (GDPS) based on gene co-expression data.
• Selected negative training samples with lower GDPS to improve model specificity.

Main Results:

• The HetIG-PreDiG model achieved a high prediction accuracy (Micro-F1 = 0.95) for gene-disease associations.
• Outperformed baseline models in predicting novel disease genes.
• Model predictions were validated using published literature.

Conclusions:

• HetIG-PreDiG effectively overcomes limitations of previous graph-based methods for disease gene prediction.
• The model advances the understanding of complex genetic diseases by accurately identifying novel disease-associated genes.
• This approach provides a powerful tool for genomic research and personalized medicine.