Subclonal reconstruction of tumors by using machine learning and population genetics

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Most cancer genomic data comes from bulk samples, mixing cancer subpopulations and normal cells.
• Current subclonal reconstruction methods are data-driven and ignore evolutionary principles, leading to errors.

Purpose of the Study:

• To develop a model-based approach for tumor subclonal reconstruction that incorporates evolutionary theory.
• To improve the accuracy and robustness of inferring cancer evolutionary history.

Main Methods:

• Developed MOBSTER, a novel method combining machine learning with theoretical population genetics.
• Validated MOBSTER using public whole-genome sequencing data (2,606 samples) and synthetic data.

Main Results:

• MOBSTER demonstrates superior robustness and accuracy compared to existing methods.
• The model-based approach minimizes errors caused by ignoring evolutionary dynamics.
• Accurate recovery of human cancer evolutionary history is achieved across single-sample, multi-region, and longitudinal data.

Conclusions:

• Integrating evolutionary theory into machine learning models significantly enhances subclonal reconstruction.
• MOBSTER offers a more reliable tool for understanding cancer evolution and informing treatment strategies.