Sensitivity analysis-guided model reduction of a mathematical model of pembrolizumab therapy for de novo metastatic MSI-H/dMMR colorectal cancer

TL;DR

This paper presents a sensitivity analysis-guided reduction of a detailed mathematical model of pembrolizumab therapy in metastatic MSI-H/dMMR colorectal cancer, resulting in simplified yet accurate models for future research.

Contribution

The authors developed two simplified models of the immune response in MSI-H/dMMR colorectal cancer using sensitivity analysis, enhancing model tractability and extensibility.

Findings

Two models accurately replicate original dynamics

Simplified models facilitate future theoretical studies

Models preserve mechanistic interpretability

Abstract

Colorectal cancer (CRC) is the third most commonly diagnosed cancer worldwide and the leading cause of cancer-related deaths in adults under 55, involving a complex interplay of biological processes such as dendritic cell (DC) maturation and migration, T cell activation and proliferation, cytokine production, and T cell and natural killer (NK) cell-mediated cancer cell killing. Microsatellite instability-high (MSI-H) CRC and deficient mismatch repair (dMMR) CRC constitute 15% of all CRC and 4% of metastatic CRC, and exhibit remarkable responsiveness to immunotherapy, especially with PD-1 inhibitors such as pembrolizumab. Mathematical models of the underlying immunobiology and the interactions underpinning immune checkpoint blockade offer mechanistic insights into tumour--immune dynamics and provide avenues for treatment optimisation and the identification of novel therapeutic targets. We used our data-driven model of de novo metastatic MSI-H/dMMR CRC (dnmMCRC) and performed sensitivity analysis-guided model reduction using the Fourier amplitude sensitivity testing (FAST) and extended FAST (EFAST) methods. In this work, we constructed two simplified models of dnmMCRC: one that faithfully reproduces all of the original model's trajectories, and a second, minimal model that accurately replicates the original dynamics while being highly extensible for future inclusion of additional components to explore various aspects of the anti-tumour immune response. Together, these resulting models offer a tractable foundation for future theoretical and computational studies of immune checkpoint blockade, avoiding unnecessary complexity while preserving mechanistic interpretability.