# Improving Protein Structure Determination by Integrating Ensemble-Driven Molecular Dynamics with Chemical Shift-Based Restraints

## TL;DR

This paper introduces a new method that combines molecular dynamics with chemical shift data to improve the accuracy of protein structure determination.

## Contribution

The novel protocol integrates EDMD with chemical shift-based restraints to refine protein structures more accurately.

## Key findings

- EDMD improved backbone RMSD compared to CS-Rosetta ensembles for 5 benchmark proteins.
- EDMD enhanced fulfillment of NOE-derived distance restraints over CS-Rosetta and unrestrained MD.
- EDMD maintained correct protein–ligand conformations and can refine nonconverged CS-Rosetta results.

## Abstract

We present a protocol for nuclear magnetic resonance
(NMR) chemical
shift-based structure determination that employs ensemble-driven molecular
dynamics (EDMD) for structure refinement. Here, specifically, Chemical-Shift-Rosetta
(CS-Rosetta) was applied, followed by EDMD. EDMD eliminates the need
to predict chemical shifts at every molecular dynamics (MD) step by
defining continuous, differentiable potential energy functions (PEFs)
based on dihedral angle distributions from CS-Rosetta models while
incorporating the measurement temperature. This yielded a thermodynamically
realistic, experiment-based custom force field for each studied system.
We benchmarked EDMD against 5 proteins (13.1–19.2 kDa), focusing
on systems with nonprotein components and demonstrated its consistent
improvement of backbone root-mean-square deviation (RMSD) relative
to known reference structures over the original CS-Rosetta ensemble.
Moreover, EDMD enhanced the fulfillment of NOE-derived (nuclear Overhauser
effect) distance restraints compared to the results of CS-Rosetta
and unrestrained MD simulations. EDMD also improved NOE-RASREC-Rosetta
(resolution-adapted structural recombination Rosetta protocol supplemented
with NOE-based distance restraints) models and maintained the correct
protein–ligand conformations. This approach provides an opportunity
to refine nonconverged CS-Rosetta structure calculations, where the
results would not be interpretable otherwise. EDMD can be generalized
to any ensemble with scoring information, enabling refined exploration
of the φ/ψ phase space and accurate reinsertion of nonprotein
moieties.