Multiplex networks-based directed graph neural network for cancer driver gene identification.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Identifying cancer driver genes is vital for precision oncology.
• Current methods often use single networks, limiting comprehensive gene relationship analysis.
• Integrating multiplex networks with multi-omics data offers a richer view of gene interactions.

Purpose of the Study:

• To develop a novel method for identifying cancer driver genes.
• To address limitations of existing methods, including label scarcity and class imbalance.
• To leverage multiplex networks and multi-omics data for improved gene representation.

Main Methods:

• Proposed a multiplex networks-based directed graph neural network (MNDGNN).
• Utilized directed graph convolution to integrate neighbor and degree diversity from multiplex networks.
• Incorporated data augmentation (positive-sample augmentation and negative-sample inference) to mitigate label scarcity.

Main Results:

• MNDGNN demonstrated superior predictive performance and robustness compared to state-of-the-art methods.
• Predicted cancer driver genes were significantly enriched in cancer-related pathways.
• Identified extensive interactions between predicted genes and known cancer drivers.

Conclusions:

• The MNDGNN model effectively identifies cancer driver genes by integrating multiplex networks and multi-omics data.
• The approach offers a new perspective for cancer driver gene discovery.
• Findings support the design of novel therapeutic strategies in precision oncology.