Every which way? On predicting tumor evolution using cancer progression models.

Area of Science:

• Computational biology
• Evolutionary medicine
• Cancer research

Background:

• Predicting tumor progression paths is crucial for cancer diagnosis, prognosis, and treatment.
• Cancer progression models (CPMs) infer tumor evolutionary trajectories from cross-sectional data by identifying mutation accumulation order.

Purpose of the Study:

• To evaluate the performance of four CPMs in predicting tumor progression paths and evolutionary unpredictability.
• To assess the impact of sample size, fitness landscape complexity, and detection regime on CPM accuracy.

Main Methods:

• Simulations were used to test CPM performance under varying conditions (sample size, fitness landscapes, detection regimes).
• Four distinct CPMs were analyzed for their ability to predict true progression path distributions.
• Real-world cancer datasets (n=22) were analyzed to assess evolutionary unpredictability and prediction reliability.

Main Results:

• CPMs perform well with large sample sizes on single-peaked fitness landscapes but poorly on multi-peaked landscapes.
• Performance is consistently suboptimal with small sample sizes common in current cancer studies.
• Estimates of evolutionary unpredictability from CPMs tend to overestimate true unpredictability, with bias influenced by the detection regime.

Conclusions:

• While CPMs offer potential for cancer progression prediction, current methods yield unreliable path predictions, especially with increasing features.
• Methodological advancements are needed to accurately model complex, multi-peaked fitness landscapes characteristic of cancer evolution.
• CPMs may serve as tools for estimating upper bounds of evolutionary unpredictability, but practical utility requires further development.