Abstract
The identifiability analysis of linear Ordinary Differential Equation (ODE)systems is a necessary prerequisite for making reliable causal inferences aboutthese systems. While identifiability has been well studied in scenarios wherethe system is fully observable, the conditions for identifiability remainunexplored when latent variables interact with the system. This paper aims toaddress this gap by presenting a systematic analysis of identifiability inlinear ODE systems incorporating hidden confounders. Specifically, weinvestigate two cases of such systems. In the first case, latent confoundersexhibit no causal relationships, yet their evolution adheres to specificfunctional forms, such as polynomial functions of time $t$. Subsequently, weextend this analysis to encompass scenarios where hidden confounders exhibitcausal dependencies, with the causal structure of latent variables described bya Directed Acyclic Graph (DAG). The second case represents a more intricatevariation of the first case, prompting a more comprehensive identifiabilityanalysis. Accordingly, we conduct detailed identifiability analyses of thesecond system under various observation conditions, including both continuousand discrete observations from single or multiple trajectories. To validate ourtheoretical results, we perform a series of simulations, which support andsubstantiate our findings.