The heterogeneous conformational ensembles of flexible proteins are critical to their function in bacterial virulence, cellular regulation, and other physiological roles, yet most high-resolution structural methods succeed precisely by trapping such proteins in one or two conformational states. Spectroscopic techniques such as DEER and single-molecule FRET can yield population distributions for conformational ensembles but do this for a small set of label-label distances. The driving hypothesis of this project is that advanced computational models can be used not only to help refine protein conformational ensembles using DEER or smFRET data but also to direct such experiments by predicting the optimal placement of labels. We therefore develop new methods 1) to predict the most informative placement for a set of labels given an initial, incomplete estimate of a conformational ensemble and 2) to better refine conformational ensembles given heterogeneous conformational population data, potentially from multiple experimental sources acquired under different conditions. We apply these methods to gain insight into flexible molecular recognition and cellular invasion by Neisseria and into cellular regulation of the synaptic fusion machinery that controls neurotransmission.