Automated Force-Field Parametrization Guided by Multisystem Ensemble Averages

RNA structure and dynamics play a fundamental role in many cellular processes such as gene expression inhibition, splicing and catalysis. Molecular dynamics is a computational tool that can be used to investigate RNA structure and dynamics at atomistic resolution. However, its capability to predict and explain experimental data is limited by the accuracy of the employed potential energy functions, also known as force fields. Recent works have shown that state-of-the-art force fields could predict unphysical RNA conformations that are not in agreement with experiments. The emerging strategy to overcome these limitations is to complement molecular dynamics with experimental data included as restraints. In a recent work, we suggested a maximum-entropy based method to enforce solution experiments in molecular dynamics simulations. Importantly, this approach reduces the risk of overfitting by simultaneously adapting the force-field corrections to multiple systems. We here push this idea further and develop a general scheme to fit arbitrary force-field parameters given a set of ensemble averages. Such quantities can range from NMR data, such as 3J couplings or NOE, to native state populations. The key feature of our method is the possibility to concurrently combine ensemble averages from multiple systems into a unique error function to be minimized, drastically enhancing corrections’ transferability. The method is applied to the difficult case of GAGA and UUCG tetraloops for which we are able to maximize their native state population by refining torsional potentials alone