Folding Simulations for Proteins with Diverse Topologies Are Accessible in Days with a Physics-Based Force Field and Implicit Solvent

Abstract

The millisecond time scale needed for molecular dynamics simulations to approach the quantitative study of protein folding is not yet routine. One approach to extend the simulation time scale is to perform long simulations on specialized and expensive supercomputers such as Anton. Ideally, however, folding simulations would be more economical while retaining reasonable accuracy, and provide feedback on structure, stability and function rapidly enough if partnered directly with experiment. Approaches to this problem typically involve varied compromises between accuracy, precision, and cost; the goal here is to address whether simple implicit solvent models have become sufficiently accurate for their weaknesses to be offset by their ability to rapidly provide much more precise conformational data as compared to explicit solvent. We demonstrate that our recently developed physics-based model performs well on this challenge, enabling accurate all-atom simulated folding for 16 of 17 proteins with a variety of sizes, secondary structure, and topologies. The simulations were carried out using the Amber software on inexpensive GPUs, providing ∼1 μs/day per GPU, and >2.5 ms data presented here. We also show that native conformations are preferred over misfolded structures for 14 of the 17 proteins. For the other 3, misfolded structures are thermodynamically preferred, suggesting opportunities for further improvement.

Partially Assembled Nucleosome Structures at Atomic Detail

“…Langevin dynamics was used for temperature regulation with a collision frequency (effective viscosity) g ¼ 0.01 ps -1 . The critical advantage of this low effective solvent viscosity regime, successfully used in simulation of nucleic acids, proteins, and their complexes (39,43,48,49), is that it offers~100-fold speedup of large conformational transitions relative to the more traditional explicit solvent simulations, for the types of structures considered here (44,50). For explicit solvent calculations we used the GROMACS package (51), with the AMBER ff10 force field at 300 K for 100 ns in the explicit solvent (PME, TIP3P water (52), with 0.145 M NaCl).…”

“…We use, to our knowledge, a novel approach-a combination of modern molecular modeling techniques (33)(34)(35)(36)(37)(38)(39)(40)(41)(42)(43)(44)(45)(46) and atomic force microscopy (AFM) measurements-to add experimentally consistent models of these three new members to the family of atomistic structures of the nucleosome particles. …”

Determining protein structures by combining semireliable data with atomistic physical models by Bayesian inference

“…2, which may alone be enough to correctly predict the structures of small proteins (17)(18)(19). The likelihood function, pðDjxÞ, expresses our belief about how probable it is that the observed data were produced by a particular structure (discussed below).…”

Constraint methods that accelerate free-energy simulations of biomolecules

“…46 computations-with a few exceptions. 5,6 Imagine, instead, instructing a REMD simulation to find the native structure by finding the state of lowest free energy in an atomistic simulation at the same time as "finding a good hydrophobic core" or "finding good secondary structures." The aim is to accelerate the discovery of the native structure, while preserving free energies.…”

“…[1][2][3][4][5][6][7] To design drugs or to create narratives of how the intricate mechanisms of proteins are encoded within their molecular structures often requires capturing fine spatial and temporal detail. Experimental science alone generally provides only information that is too coarse to tell the stories of the hundreds of thousands of structures and mechanisms of the protein universe.…”