The mechanical response of metals in the plastic regime is of utmost importance for structural applications. Despite significant advances in understanding the deformation mechanisms, a systematic framework for predicting plastic behavior remains elusive. Addressing this challenge requires a multiscale understanding of the collective behavior of dislocations at the mesoscale and establishing a connection with commensurate experiments. This dissertation addresses that challenge by combining advanced X-ray microscopy, machine learning, statistical analysis, and continuum dislocation dynamics to bridge microstructural observation and predictive constitutive modeling. The research in this dissertation is divided into three parts. The first part establishes a machine learning methodology for understanding structural data from high-resolution differential-aperture X-ray microscopy (HR-DAXM) experiments, which we apply to high-resolution rotation and elastic strain data for 304 steel that are obtained by collaborators. It is found that, within individual grains, the lattice rotation distribution exhibits multimodality, a signature of incipient cell formation. An unsupervised Cauchy mixture model is developed to understand these features, which deconvolves the multimodal lattice rotation distribution into distinct Cauchy components, each representing a physically meaningful rotational domain. Mapping these domains back into physical space reveals contiguous regions of lattice rotation. This interpretation is validated by comparison with the dislocation density tensor, which highlights geometrically necessary boundaries that define the emergent cell structure. The analysis demonstrates how statistical decomposition of experimental data can expose hidden features of microstructural evolution in deformed metals. The second part focuses on the collective behavior and dynamic evolution of dislocations to inform coarse-grained plasticity models. A continuum dislocation dynamics model is employed to extract physical quantities used in crystal plasticity models. This is achieved using a streamline analysis to reconstruct dislocations and their velocity fields, from which the quantification of dislocation segment lengths, free paths, and wall fraction is enabled. These descriptors establish a mechanistic foundation for mesoscale constitutive models, replacing phenomenological fitting with parameters derived directly from dislocation physics. The third part of this work employs the same framework to investigate the behavior of the microstructural parameters under complex loading conditions and with the effect of precipitates. Taken together, these investigations establish computational approaches for analyzing the dislocation microstructure in deformed crystals, both for the purpose of informing plasticity models at the macroscale and analysis of relevant mesoscale experiments.