SSCI: Self-Supervised Deep Learning Improves Network Structure for Cancer Driver Gene Identification.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Cancer pathogenesis involves genetic abnormalities.
• Accurate identification of cancer genes is critical for early detection and personalized medicine.
• Graph deep learning methods show promise for identifying cancer driver genes from biological networks, but network noise and incompleteness hinder performance.

Purpose of the Study:

• To propose a novel method for cancer driver gene identification.
• To address the limitations of existing graph deep learning methods caused by noisy and incomplete biological networks.
• To enhance biological network structures and improve predictive accuracy using self-supervision.

Main Methods:

• Development of a self-supervised learning framework for graph convolutional networks (GCNs).
• Application of the proposed method, Self-Supervised Cancer Gene Identification (SSCI), to enhance network structure.
• Evaluation of SSCI's performance using standard metrics: Area Under the Receiver Operating Characteristic Curves (AUROC), Area Under the Precision-Recall Curves (AUPRC), and F1 score.

Main Results:

• The SSCI method achieved high reliability, with AUROC of 0.966, AUPRC of 0.964, and F1 score of 0.913.
• The proposed approach effectively enhances biological network structures.
• The method demonstrates strong discriminative power in identifying cancer driver genes.

Conclusions:

• The developed self-supervised method significantly improves cancer driver gene identification.
• SSCI offers enhanced biological network representation for more accurate predictions.
• The findings suggest that SSCI has strong discriminative power and biological interpretability for cancer gene discovery.