Trauma is the leading cause of death and disability in United States for both children and adults. In response to trauma, the body unleashes a set of coupled programs that affect the functioning of vascular, immune and autonomous nervous systems. In pathological cases, the integrated output of these programs can result in coagulopathy, systemic in- flammatory response syndrome (SIRS), multiple organ dysfunction syndrome (MODS) and potentially even death. Nearly 35%-40% of trauma deaths occur due to uncontrolled hemorrhage resulting from trauma-induced coagulopathy (TIC). TIC also plays an impor- tant role in modulating inflammation, organ dysfunction and increased susceptibility to sepsis. Clinical trials for treatment strategies targeting TIC have met with limited success. The interlinked nature of coagulant and inflammatory responses, along with patient spe- cific physiological variability, make the treatment of TIC challenging. Understanding TIC requires an integrated multi-scale modeling framework which describes the relevant bio- chemical networks within the context of the whole-body. Given the complexity and size, embedding large, non-linear models of biochemical networks into a whole body model creates a significant computational challenge. Thus an objective of this work is to develop a framework that reduces the complexity of high-dimensional mathematical models. We apply this framework to model biochem- ical networks that are important in TIC. We first investigate the dynamics of coagulation and understand the impact of protein C pathway on thrombin generation. Thereafter we use this reduced order modeling technique to model complement and fibrinolysis. We identify targets of therapeutic importance in complement and mechanisms that control clot degradation in fibrinolysis. We show that we can capture the dynamics of these com- plex but varied systems using the reduced order modeling framework. In addition, we address the problem of training high-dimensional, non-linear models of biological systems. Traditional gradient based methods often fail due to convergence to a local optima or due to the lack of gradient knowledge. We present a novel optimiza- tion method that is based on evolutionary algorithms to obtain near optimal parameters within a limited number of function evaluations. We demonstrate that this method ob- tains optimal solutions on a wide array of non-linear models, faster than existing meta heuristic methods. Taken together this work provides a methodology to rapidly investi- gate complex biochemical systems by simplifying the model design and experimentation processes.