Abstract
Reliable genome-scale metabolic models (GEMs) of metabolic processes are important for understanding cellular behavior. However, the presence of thermodynamically infeasible cycles (TICs) limits their predictive ability. We present ThermOptCOBRA, a comprehensive solution consisting of four algorithms for optimal model construction and analysis that integrate thermodynamic constraints to address TICs. By leveraging network topology, ThermOptCOBRA efficiently identifies TICs in 7,401 published models. It determines thermodynamically feasible flux directions, thereby detecting the blocked reactions, which yields more refined models with fewer TICs. Furthermore, it constructs thermodynamically consistent context-specific models that are compact in comparison to Fastcore in 80% of cases. ThermOptCOBRA also facilitates efficient loop detection and removal from flux distributions, improving predictive accuracy across flux analysis methods. Moreover, it enhances sampling algorithms by enabling loopless sample generation. In summary, ThermOptCOBRA significantly improves TIC handling in GEMs, advancing metabolic model quality for deeper insights into cellular metabolism.