Biophysical techniques such as electron paramagnetic resonance (EPR) and hydrogen-deuterium exchange mass spectrometry (HDX-MS) provide powerful insights into biomolecular structure and dynamics. While each technique yields ensemble-averaged observables of relatively low resolution, combining multiple complementary data sets can uncover structural information often missed by high-resolution techniques or structure prediction methods, including transient but functionally important intermediates. Extracting this information requires integration with computational models that rigorously account for both measurement uncertainty and modeling errors. We present a software platform for ensemble refinement that integrates experimental data with molecular dynamics simulations or alternative structural models such as Rosetta or AlphaFold. The method is grounded in a maximum entropy framework, introducing minimal bias to the initial ensemble while enforcing consistency with experimental observables within their uncertainties. The platform can accommodate both raw experimental signals and derived quantities (e.g., distance distributions) and enables the simultaneous incorporation of a wide range of experimental sources. Key features include automated parameter optimization, customizable predictive models, and flexible workflows for incorporating new machine learning-based predictors. The platform supports both conventional and enhanced sampling simulations by coupling refinement with unbiasing strategies to optimize ensemble weights. Additional capabilities include automated convergence monitoring, trajectory filtering, and rigorous uncertainty quantification through effective sampling size analysis, Bayesian metrics, and cross-validation strategies. We demonstrate the methodology through applications to double electron-electron resonance (DEER) spectroscopy and HDX-MS, using both synthetic and experimental data on globular and membrane proteins. The framework is extensible and includes tailored strategies for handling experimental artifacts such as deuterium loss. By unifying complementary experimental measurements with advanced computational modeling, this platform provides a robust and versatile tool for ensemble refinement, enabling deeper molecular insights into complex biomolecular systems, while providing a rigorous framework to optimize structural and predictive models.
Byju et al. (Sun,) studied this question.