Cancer driver mutation prediction through Bayesian integration of multi-omic data

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Identifying cancer driver mutations is crucial for personalized medicine.
• Tumor heterogeneity and low-frequency mutations pose challenges for distinguishing driver from passenger mutations.

Purpose of the Study:

• To develop and validate a novel Bayesian hierarchical modeling approach (rDriver) for enhanced prediction of cancer driver mutations.
• To integrate multi-omic data, including functional impact scores and gene expression variations, for improved driver mutation identification.

Main Methods:

• Developed rDriver, a Bayesian hierarchical model integrating mutation functional impact and gene expression data.
• Applied rDriver to 3,080 tumor samples across 8 cancer types from The Cancer Genome Atlas (TCGA).
• Validated predictions using engineered cell-line models, focusing on PIK3CA genes.

Main Results:

• rDriver predicted 1,389 driver mutations across 8 cancer types.
• The model identified more low-frequency mutations linked to lineage-specific properties, occurrence timing, and patient survival compared to existing tools.
• Achieved a positive predictive value of 0.94 for PIK3CA gene predictions in cell-line models.

Conclusions:

• Integrating multi-omic data significantly enhances the accuracy of cancer driver mutation prediction.
• rDriver offers a statistically rigorous method for identifying cancer driver mutations, supporting cancer target discovery and development.
• The study underscores the value of advanced computational approaches in precision oncology.