Graph-based sufficient conditions for indistinguishability of linear compartmental models

TL;DR

This paper establishes graph-based sufficient conditions to determine when linear compartmental models are indistinguishable, enabling analysis without symbolic computation and aiding model selection in biological sciences.

Contribution

It introduces the first graph-structure-based sufficient conditions for indistinguishability in linear compartmental models, expanding beyond previous necessary conditions.

Findings

Sufficient conditions for indistinguishability based on graph paths with detours.

Indistinguishability can be verified without symbolic computation.

Models are indistinguishable up to parameter renaming (permutation indistinguishability).

Abstract

An important problem in biological modeling is choosing the right model. Given experimental data, one is supposed to find the best mathematical representation to describe the real-world phenomena. However, there may not be a unique model representing that real-world phenomena. Two distinct models could yield the same exact dynamics. In this case, these models are called indistinguishable. In this work, we consider the indistinguishability problem for linear compartmental models, which are used in many areas, such as pharmacokinetics, physiology, cell biology, toxicology, and ecology. We exhibit sufficient conditions for indistinguishability for models with a certain graph structure: paths from input to output with "detours". The benefit of applying our results is that indistinguishability can be proven using only the graph structure of the models, without the use of any symbolic computation. This can be very helpful for medium-to-large sized linear compartmental models. These are the first sufficient conditions for indistinguishability of linear compartmental models based on graph structure alone, as previously only necessary conditions for indistinguishability of linear compartmental models existed based on graph structure alone. We prove our results by showing that the indistinguishable models are the same up to a renaming of parameters, which we call permutation indistinguishability.