Foundation models for atomic interactions are now within reach, opening new possibilities for atomistic simulation of materials and chemistry, and driving progress in materials modeling. However, foundation models that can capture microstructural features within materials are still unavailable as they span larger length and time scales than atomic interactions. Accurately modeling such features requires upscaling or scale-bridging to map atomistic behavior to larger microstructural phenomena like dislocations. Dislocations are crystal defects that enable metals to undergo plastic deformation. A prerequisite for upscaling atomistic interactions to larger scale phenomena requires datasets detailing dislocation network evolution, which are currently unavailable at the scale required. A key step forward will be the creation of a representative set of large-scale dislocation trajectories based on atomistic simulation. However, it is yet unclear which kind of boundary conditions, e.g. through appropriate thermostatting yield the correct dislocation behavior in atomistic simulation. In this study, we systematically investigate which thermostat yields physically accurate deformation mechanisms in large-scale molecular dynamics simulations, especially under high strain rates and varying parameters such as damping coefficients. Once realistic dislocation networks are obtained, dislocation trajectories are analyzed through graph representations and dislocation density fields to predict which are then linked to macroscopic mechanical properties. This work represents a first step towards data-based view on plastic deformation based on atomistic simulations.