Associating somatic mutation with clinical outcomes through kernel regression and optimal transport

Area of Science:

• Oncology
• Computational Biology
• Genomics

Background:

• Somatic mutations in cancer are sparse and high-dimensional, making gene-by-gene analysis underpowered.
• Cancer patients share deregulated biological processes, even with different mutated genes.
• Current methods often fail to capture complex, shared disease mechanisms.

Purpose of the Study:

• To develop a principled method for aggregating somatic mutation data to assess joint associations with clinical outcomes.
• To account for gene-gene similarities derived from annotations or mutational patterns.
• To overcome the limitations of gene-by-gene analyses in cancer genomics.

Main Methods:

• Utilized optimal transport to estimate somatic mutation profile similarity between tumor samples.
• Incorporated gene-gene similarities based on annotations or empirical mutational patterns.
• Applied kernel regression to assess associations between aggregated mutation profiles and clinical outcomes.

Main Results:

• Successfully analyzed somatic mutation data across 17 cancer types.
• Identified significant associations between somatic mutations and clinical outcomes in at least five cancer types.
• Demonstrated the utility of the method in linking mutation profiles to overall survival, progression-free interval, and cytolytic activity.

Conclusions:

• The proposed optimal transport-based approach effectively aggregates sparse, high-dimensional somatic mutation data.
• This method provides a powerful framework for uncovering complex relationships between cancer mutations and clinical outcomes.
• The findings highlight the importance of considering joint mutation effects for a comprehensive understanding of cancer biology and patient prognosis.