Detecting repeated cancer evolution from multi-region tumor sequencing data.

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Genomic changes in cancer reflect evolutionary processes crucial for predicting disease progression.
• While multi-region sequencing aids in inferring temporal genomic order within tumors, identifying consistent evolutionary patterns across patients is challenging.
• Stochasticity and data noise complicate the robust detection of convergent cancer evolution.

Purpose of the Study:

• To develop a novel machine learning method for identifying hidden evolutionary patterns in cancer cohorts.
• To overcome limitations of existing methods in detecting repeated evolution across patients.
• To enable patient classification based on tumor evolutionary trajectories for improved disease progression prediction.

Main Methods:

• Development of a transfer learning-based machine learning approach.
• Application to multi-region sequencing datasets from lung, breast, renal, and colorectal cancers (768 samples, 178 patients).
• Validation of detected evolutionary trajectories in independent single-sample cohorts (2,935 samples).

Main Results:

• The machine learning method successfully identified hidden evolutionary patterns in cancer cohorts.
• Repeated evolutionary trajectories were detected in patient subgroups across multiple cancer types.
• These findings were reproducible in larger, independent single-sample cohorts, demonstrating robustness.
• The method enabled patient classification based on distinct tumor evolutionary paths.

Conclusions:

• The developed machine learning method effectively identifies recurrent evolutionary patterns in cancer.
• This approach offers a powerful tool for classifying patients based on their tumor's evolutionary history.
• The findings have significant implications for anticipating cancer progression and guiding clinical strategies.