Prediction Of A Protein'S Free Energy Surface And Validation With Hd Exchange

Here we address a central challenge in protein biophysics, the de novo simulation of a protein’s free energy surface (FES), including the generation of the Boltzmann ensemble of the major species, followed by experimental validation. To this end, we have further improved Upside, our recently developed physics-based MD algorithm that can cooperatively fold proteins with an accuracy comparable to all-atom methods yet is 103-104 fold faster (Jumper et al., (2018) PLoS Comp Biol). Upside's energy function is improved by simultaneously training all parameters including new H-bond and desolvation terms. Correctly producing all-or-none folding cooperativity requires navigating a difficult balancing of energy terms, which we accomplish by training the forcefield with both the native and unfolded state ensembles. Hydrogen-deuterium exchange (HDX), which provides the free energies of individual H-bonds, is used to test our ability to describe the FES, including rare species such as partially folded and even unfolded states. A comparison to HDX data is quite auspicious both for site-resolved protection factors and the global stability for a variety of designed and naturally occurring proteins. Furthermore, we investigate whether Upside can correctly identify the conformations from which HDX is occurring (e.g., locally, partially or globally unfolded states). This last evaluation of Upside is carried out via a comparison of the denaturant dependence of the simulated FES to that of the HDX data. These promising results indicate that we have learned sufficient protein chemistry, and how to encode it into a model, so that we can describe a protein's free energy surface even with modest computing resources.