MODIG: integrating multi-omics and multi-dimensional gene network for cancer driver gene identification based on...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Identifying genes causally involved in cancer evolution is a significant challenge in cancer biology.
• Integrating high-throughput multi-omics data for cancer driver gene identification is complex.

Purpose of the Study:

• To propose MODIG, a graph attention network (GAT)-based framework for identifying cancer driver genes.
• To effectively integrate multi-omics pan-cancer data with multi-dimensional gene networks.

Main Methods:

• Constructed a multi-dimensional gene network with ~20,000 genes and five types of associations.
• Applied a GAT to model within-dimension interactions and generate gene representations.
• Developed a joint learning module to fuse dimension-specific representations for general gene representations.
• Utilized the generated gene representations for semi-supervised driver gene identification.

Main Results:

• MODIG integrates mutations, copy number variants, gene expression, and methylation data.
• The framework leverages diverse gene relationship maps (PPI, sequence similarity, KEGG, co-expression, GO).
• MODIG demonstrates superior performance compared to baseline models in driver gene identification tasks, evidenced by higher area under precision-recall and receiver operating characteristic curves.

Conclusions:

• MODIG provides an effective computational framework for identifying cancer driver genes.
• The integration of multi-omics data and multi-dimensional gene networks enhances driver gene discovery.
• The developed GAT-based approach offers a promising direction for advancing cancer genomics research.