Integrated structural variation and point mutation signatures in cancer genomes using correlated topic models

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Cancer genomes contain mutation signatures revealing mutational processes.
• These signatures offer insights into tumor etiology, prognosis, and therapeutic vulnerabilities.

Purpose of the Study:

• To develop an improved machine learning method for inferring cancer mutation signatures.
• To integrate single nucleotide and structural variation data for enhanced signature discovery.

Main Methods:

• A novel machine learning approach using multi-modal correlated topic models (MMCTM).
• Application of MMCTM to hormone-driven, DNA repair-deficient breast and ovarian cancers (755 samples).

Main Results:

• MMCTM enhances signature discovery accuracy, especially with sparse data.
• Incorporating correlated structures within and between mutation modes improves inference.

Conclusions:

• Integrating multiple mutation modes is crucial for accurate cancer signature discovery and patient stratification.
• The developed statistical framework supports future inclusion of additional biological features.