Computational Methodology

The MCMC simulations were carried out using PHAISTOS with the Metropolis–Hastings algorithm, using the energy function described above.The conformational degrees of freedom explored in the simulations were restricted to the backbone and side-chain dihedral angles (φ,ψ,χ) as well as the backbone bond angles.

Two types of simulations are performed: a constant temperature simulations designed to generate an ensemble that best reflects the fast dynamics of the protein using the hybrid energy function and a simulated annealing simulation that seeks to find the lowest energy structure.

8 independent Metropolis-Hastings simulations were run for 50 M MC steps each at 300 K. The physical move set was comprised of 20% uniform and 30% local single side chain moves, 40% CRISP local backbone dihedral angle moves and 10% CRA backbone bond angle moves. An additional 5% moves were added to sample the standard deviation in the chemical shift energy term [ref weight paper].

Nine protein structures are used in this study: 3CWI, Protein G (2OED), 3E0E, 1LFO, ubiquitin (1UBQ), Prospero Homeobox protein 1 (2LMD), 3LD7, 1VC1, 2MBL. 2LMD and 2MBL are NMR ensembles while the rest are crystal structures. The starting structures for the protein refinements are constructed as follows. Protons are added to the x-ray structure by REDUCE [ref], while the median structure with respect to all atom rmsd is selected from the NMR ensembles. The structures are minimized with the CHARMM22/CMAP force field and the GB/SA solvation model implemented in TINKER, with a convergence criterion of 0.01.