Improving cancer driver gene identification using multi-task learning on graph convolutional network.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer arises from accumulated genetic mutations.
• Identifying cancer driver genes is crucial for understanding cancer mechanisms.
• Driver genes are essential for developing targeted therapies and biomarkers.

Purpose of the Study:

• To propose a novel Multi-Task learning method (MTGCN) for identifying cancer driver genes.
• To leverage Graph Convolutional Networks and protein-protein interaction (PPI) networks for enhanced gene feature representation.
• To simultaneously optimize node and link prediction tasks for robust driver gene identification.

Main Methods:

• Developed MTGCN, a Multi-Task learning framework based on Graph Convolutional Networks.
• Augmented gene features using their relationships within the protein-protein interaction (PPI) network.
• Employed a Bayesian task weight learner to automatically balance node and link prediction tasks.

Main Results:

• MTGCN assigns a probability score for each gene being a cancer driver.
• The model demonstrated superior performance compared to four existing methods.
• Evaluated performance using Area Under the ROC Curve and Area Under the Precision-Recall Curve for pan-cancer and single-cancer types.

Conclusions:

• MTGCN effectively identifies cancer driver genes.
• The proposed method outperforms state-of-the-art approaches.
• MTGCN offers a valuable tool for cancer research and precision medicine development.