ORBIT: Oncogenic Representation Learning via Bi-Prototype Contrastive Learning in Hyperbolic Space for cancer driver...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Accurate identification of cancer driver genes is essential for advancing precision oncology.
• Integrating heterogeneous multi-omics data and modeling dynamic biological systems presents significant challenges.

Purpose of the Study:

• To develop a novel computational framework, ORBIT, for robust and accurate identification of cancer driver genes.
• To address limitations in current methods by integrating multi-omics profiles and functional network data.

Main Methods:

• ORBIT utilizes a context-adaptive graph reweighting mechanism to fuse multi-omics and network data, capturing cancer-specific dynamics.
• A bi-prototype contrastive learning strategy in hyperbolic space aligns gene representations around driver and non-driver semantic anchors.
• The approach preserves the intrinsic hierarchy of biological networks while modeling dynamic gene interactions.

Main Results:

• ORBIT demonstrates high stability in pan-cancer analyses and outperforms state-of-the-art methods in cancer-specific predictions.
• Functional enrichment analysis confirms ORBIT's ability to segregate core cancer pathways.
• Drug sensitivity profiling validates the clinical relevance of the driver genes identified by ORBIT.

Conclusions:

• ORBIT offers a robust and interpretable paradigm for cancer driver gene identification by integrating hyperbolic geometry and context-adaptive learning.
• The framework advances precision medicine by providing more accurate and clinically relevant driver gene predictions.
• The developed model and associated datasets are publicly available to facilitate further research.