Coarse Graining to Expedite Molecular Dynamics Simulations of Solvated Fibrinogen

Accurate molecular dynamics (MD) simulations often require massive computing resources to simulate millions to billions of individual atoms. These all-atomic (AA) simulations are computationally intensive, and often fail to reach long simulation time scales [1]. Coarse graining (CG) alleviates these limitations by reducing an all-atomic model to a skeleton with fewer particles while preserving accuracy by ensuring that the interparticle forces closely resemble those of the all-atomic model [1]. Fibrinogen is especially important in modern biology due to its ability to form fibrin, a protein central to blood clotting and controlling bleeding in vascular injury. A CG model of Fibrinogen with alpha C domains in a vacuum was created by [2] in 2021 using machine learning methods. Our research builds upon the CG model of [2] by solvating it with water to more accurately emulate Fibrinogen’s naturally solvated form. Such a solvated model would greatly boost pharmaceutical research in several important areas including thrombosis prevention.<br/><br/>To ensure that our CG model of solvated Fibrinogen accurately emulates an all-atomic model (AA model), we create a control system with the all-atomic 3GHG PDB file. We utilize the structure and force files from [2] for the CG Fibrinogen model. VMD’s water box creation tool is employed to solvate both systems. The AA model is solvated with TIP3 water, whereas the CG model is solvated with MARTINI water with the same water box dimensions. We undertake NVT simulations at 310K with Langevin dynamics. To get comparable analyses from the AA model, we group atoms of the resultant AA trajectory into CG beads in the same way the original CG model was formed in [2], referred to as the “mapped all atomic” (MAA) model. For accuracy analysis, we measure the MAA and CG models' root-mean-squared deviation (RMSD), free energy distributions as a function of dihedral angles, and radial distribution functions (RDF).<br/><br/>Our prototyping CG system possesses 85,861 atoms including 15 atoms for Fibrinogen, whereas the AA system consists of 1,186,598 atoms including 42,003 atoms for Fibrinogen in atomistic resolution. On a 2GHz Quad-Core Intel Core i5, the CG model took 41 seconds of wall clock time to complete a 1 picosecond simulated time, while the AA model took 7028 seconds––the CG simulation achieving 172x speedup. After conducting 14.16 nanoseconds of CG simulation with 10 femtoseconds as the time step size and 249 picoseconds AA simulation with 1 femtosecond as the time step size, we conduct our accuracy analyses. The RMSD plot of the MAA model is steeply increasing, matching almost exactly with the CG plot. Because the CG simulation is significantly less computationally expensive, we could continue it beyond the results of the AA simulation until it plateaus at an RMSD value of around 0.37 Å. The dihedral plots of the MAA and CG models all appear very similar, with higher concentrations in similar areas. Likewise, the RDF plots demonstrate highly similar distributions between the two models.<br/><br/>We conclude that our CG model of solvated Fibrinogen utilizes much less CPU time than the AA model while emulating most of the structure properties, meaning that our CG model will feasibly and accurately achieve significantly longer simulation times than the AA model. This will boost future research on Fibrinogen in essential pharmaceuticals and other essential biological fields.<br/><br/><b>Acknowledgments:</b><br/>We thank the Stony Brook University Garcia Center for Materials Research, the Garcia Summer Scholars Program, and the Morin Charitable Trust for supporting and facilitating this research.<br/><br/><b>References:</b><br/>[1] Husic, B. E., Charron, N. E., et al. (2020). Coarse graining molecular dynamics with graph neural networks. The Journal of chemical physics, 153(19):194101<br/>[2] Zhang, Z., Zhang, P.,et al. (2021). Ai-guided multiscale biomechanical model of fibrinogen: correlating with in vitro results. International Fibrinogen Research Society Workshop