DriverMONI: Cancer Driver Gene Prediction With Multimodal Deep Learning Integrating Multiomics Data and...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Driver gene identification is crucial for understanding cancer.
• Existing methods, including graph neural networks, face challenges due to static and incomplete biological networks.

Purpose of the Study:

• To introduce DriverMONI, a novel multimodal approach for accurate driver gene prediction.
• To leverage complementary information from multiomics data and biological networks.

Main Methods:

• DriverMONI uses condition-specific protein-protein interaction subnetworks to generate input graphs.
• It employs a graph attention network with node attributes for condition-specific predictions.
• The approach integrates multiomics data with network information.

Main Results:

• DriverMONI demonstrates the importance of multimodality in driver gene prediction.
• The method effectively mitigates issues arising from incomplete protein-protein interaction networks.
• Comparative analysis on The Cancer Genome Atlas data shows DriverMONI outperforms existing methods, including graph neural network-based models.

Conclusions:

• DriverMONI offers a robust and accurate solution for driver gene identification.
• The multimodal approach enhances predictive power by combining diverse biological data.
• The developed tool shows strong consensus with other methods, validating its performance.