Thermodynamics imposes ultimate restrictions on how machines can operate, but living cells have to abide by the same laws. Cellular metablism is the system by which cells extract useful energy from their food. It is composed of many interacting enzymes that form a large chemical reaction network. The understanding of such networks is therefore crucial for the understanding of life itself. In the living cell, some of these chemical processes operate naturally in a very noisy environment and many of them far from equilibrium. Modern methods from Stochastic Thermodynamics and nearby disciplines allow us to properly formulate the thermodynamics of such driven systems in a noisy environment. Investigating the thermodynamics of living cells can serve as a guiding principle to understand their functioning.In this project I focused mainly on two issues: fluctuations and coarse graining.Recent developments proved that in stochastic systems there is a close interplay between network topology and fluctuations. I investigated how the peculiar structure of metabolic networks and their fluctuations beyond equilibrium are connected. In one special topological class, the fluctuations are restricted so much that they are essentially trivial (Poissonian) extensions of deterministically calculated averages.Metabolic networks are huge, which makes their analysis very challenging. A typical simplification step is to use coarse-grained kinetics for enzymes instead of keeping track of all the details of their catalytic mechanisms. I investigated the thermodynamic consistency in the common coarse-graining approaches and modified them in such a way that the coarse-grained kinetics captures all of the essential non-equilibirum features. In doing so I also extended the applicability of this coarse-graining to more complex catalysts, such as molecular motors or active pumps.My research layed the foundations for a more rigorous treatment of thermodynamics in biochemical reaction networks.