The trade-off between parsimony and model complexity for understanding biomedical mechanisms from mathematical models

Mechanistic mathematical models are widely employed to enhance comprehension of biological processes, such as tumor growth and its reaction to different therapies. These models are often intricate and susceptible to challenges in accurately estimating parameters. To address this, statistical tools like Akaike and Bayesian information criteria are utilized to strike a balance between model simplicity and overall performance.

However, overly simplistic models might not offer adequate biological understanding if they do not fully encapsulate established physiological processes or mechanisms. Consequently, researchers must find equilibrium between model generation, biological learning, and tractability. This article showcases this delicate balance by examining ovarian cancer growth and treatment response to cisplatin and immune checkpoint blockade in genetically modified mouse models—a combination of HR-deficient and HR-proficient immunocompetent mice.

The researchers construct a series of mathematical models, ranging from simple to complex, to represent tumor growth, treatment response, and immune system dynamics. The findings emphasize the constraints of depending exclusively on statistical metrics for model selection, particularly when the focus is on gaining biological insights.

Moreover, the study underscores the significance of balancing model complexity to prevent overfitting and parameter identifiability issues.