Determining parameter impact in systems biology models via sensitivity analysis: a comparative approach.

Abstract

Mathematical modeling is a powerful tool to understand biological phenomena, predict behavior, and guide experiments. Model parameters represent experimentally derived rates, and it can be challenging to understand how parameters affect the system's output(s). Global sensitivity analysis (GSA) can help determine how uncertainty in model outcomes can be attributed to parameters. Methods like Sobol' indices provide detailed analysis but at a high computational cost. Thus, it may be necessary to precede it by another method to reduce the parameter space. Morris Method Screening (MMS) is a qualitative analysis that can provide information about sensitivity and parameter interactions, making it a good partner for Sobol'. Partial Rank Correlation Coefficient (PRCC) may also be a viable option as it has a similar cost to MMS. This work tests whether a ranking (PRCC) or screening (MMS) method is better when used in combination with Sobol' analysis. Using a model of the epithelial-mesenchymal transition as a test case, this work shows that PRCC-Sobol' and MMS-Sobol' can produce diverging results, making it difficult to analyze how biological events are informed by specific rates and quantities. However, the differences between these methods mean that these techniques may be most effective when all three are used together.