Machine Learning-Driven Multiscale Modeling: Bridging the Scales with a Next-Generation Simulation Infrastructure.

Interdependence across time and length scales is common in biology, where atomic interactions can impact larger-scale phenomenon. Such dependence is especially true for a well-known cancer signaling pathway, where the membrane-bound RAS protein binds an effector protein called RAF. To capture the driving forces that bring RAS and RAF (represented as two domains, RBD and CRD) together on the plasma membrane, simulations with the ability to calculate atomic detail while having long time and large length- scales are needed. The Multiscale Machine-Learned Modeling Infrastructure (MuMMI) is able to resolve RAS/RAF protein-membrane interactions that identify specific lipid-protein fingerprints that enhance protein orientations viable for effector binding. MuMMI is a fully automated, ensemble-based multiscale approach connecting three resolution scales: (1) the coarsest scale is a continuum model able to simulate milliseconds of time for a 1 µm2 membrane, (2) the middle scale is a coarse-grained (CG) Martini bead model to explore protein-lipid interactions, and (3) the finest scale is an all-atom (AA) model capturing specific interactions between lipids and proteins. MuMMI dynamically couples adjacent scales in a pairwise manner using machine learning (ML). The dynamic coupling allows for better sampling of the refined scale from the adjacent coarse scale (forward) and on-the-fly feedback to improve the fidelity of the coarser scale from the adjacent refined scale (backward). MuMMI operates efficiently at any scale, from a few compute nodes to the largest supercomputers in the world, and is generalizable to simulate different systems. As computing resources continue to increase and multiscale methods continue to advance, fully automated multiscale simulations (like MuMMI) will be commonly used to address complex science questions.

Machine learning-driven multiscale modeling reveals lipid-dependent dynamics of RAS signaling proteins.

RAS is a signaling protein associated with the cell membrane that is mutated in up to 30% of human cancers. RAS signaling has been proposed to be regulated by dynamic heterogeneity of the cell membrane. Investigating such a mechanism requires near-atomistic detail at macroscopic temporal and spatial scales, which is not possible with conventional computational or experimental techniques. We demonstrate here a multiscale simulation infrastructure that uses machine learning to create a scale-bridging ensemble of over 100,000 simulations of active wild-type KRAS on a complex, asymmetric membrane. Initialized and validated with experimental data (including a new structure of active wild-type KRAS), these simulations represent a substantial advance in the ability to characterize RAS-membrane biology. We report distinctive patterns of local lipid composition that correlate with interfacially promiscuous RAS multimerization. These lipid fingerprints are coupled to RAS dynamics, predicted to influence effector binding, and therefore may be a mechanism for regulating cell signaling cascades.

Capturing Biologically Complex Tissue-Specific Membranes at Different Levels of Compositional Complexity.

Plasma membranes (PMs) contain hundreds of different lipid species that contribute differently to overall bilayer properties. By modulation of these properties, membrane protein function can be affected. Furthermore, inhomogeneous lipid mixing and domains of lipid enrichment/depletion can sort proteins and provide optimal local environments. Recent coarse-grained (CG) Martini molecular dynamics efforts have provided glimpses into lipid organization of different PMs: an "Average" and a "Brain" PM. Their high complexity and large size require long simulations (∼80 µs) for proper sampling. Thus, these simulations are computationally taxing. This level of complexity is beyond the possibilities of all-atom simulations, raising the question-what complexity is needed for "realistic" bilayer properties? We constructed CG Martini PM models of varying complexity (63 down to 8 different lipids). Lipid tail saturations and headgroup combinations were kept as consistent as possible for the "tissues'" (Average/Brain) at three levels of compositional complexity. For each system, we analyzed membrane properties to evaluate which features can be retained at lower complexity and validate eight-component bilayers that can act as reliable mimetics for Average or Brain PMs. Systems of reduced complexity deliver a more robust and malleable tool for computational membrane studies and allow for equivalent all-atom simulations and experiments.

High-frequency response of classical strongly coupled plasmas.

The dynamic structure factor (DSF) of the Yukawa system is here obtained with highly converged molecular dynamics (MD) over the entire liquid phase. The data provide a rigorous test of theoretical models of ion-acoustic wave-dispersion relations, the intermediate scattering function, and the high-frequency response. We compare our MD results with seven diverse models, finding good agreement among those that enforce the three basic sum rules for dispersion properties, although one of the models has previously unreported spurious peaks. The MD simulations reveal that at intermediate frequencies ω, the high-frequency response of the DSF follows a power law, going approximately as ω^{-p}, where p>0, and p shows nontrivial dependencies on the wave vector q and the plasma parameters κ and Γ. In contrast, among the seven comparison models, the predicted high-frequency response is found to be independent of {q,κ,Γ}. This high-frequency power suggests a useful fitting form. In addition, these results expose limitations of several models and, moreover, suggest that some approaches are difficult or impossible to extend because of the lack of finite moments. We also find the double-plasmon resonance peak in our MD simulations that none of the theoretical models predicts.

A generalized Ewald decomposition for screened Coulomb interactions.

Medium-range interactions occur in a wide range of systems, including charged-particle systems with varying screening lengths. We generalize the Ewald method to charged systems described by interactions involving an arbitrary dielectric response function ϵ(𝐤). We provide an error estimate and optimize the generalization to find the break-even parameters that separate a neighbor list-only algorithm from the particle-particle particle-mesh algorithm. We examine the implications of different choices of the screening length for the computational cost of computing the dynamic structure factor. We then use our new method in molecular dynamics simulations to compute the dynamic structure factor for a model plasma system and examine the wave-dispersion properties of this system.

Atomic bound state and scattering properties of effective momentum-dependent potentials.

Effective classical dynamics provide a potentially powerful avenue for modeling large-scale dynamical quantum systems. We have examined the accuracy of a Hamiltonian-based approach that employs effective momentum-dependent potentials (MDPs) within a molecular-dynamics framework through studies of atomic ground states, excited states, ionization energies, and scattering properties of continuum states. Working exclusively with the Kirschbaum-Wilets (KW) formulation with empirical MDPs [C. L. Kirschbaum and L. Wilets, Phys. Rev. A 21, 834 (1980)0556-279110.1103/PhysRevA.21.834], optimization leads to very accurate ground-state energies for several elements (e.g., N, F, Ne, Al, S, Ar, and Ca) relative to Hartree-Fock values. The KW MDP parameters obtained are found to be correlated, thereby revealing some degree of transferability in the empirically determined parameters. We have studied excited-state orbits of electron-ion pair to analyze the consequences of the MDP on the classical Coulomb catastrophe. From the optimized ground-state energies, we find that the experimental first- and second-ionization energies are fairly well predicted. Finally, electron-ion scattering was examined by comparing the predicted momentum transfer cross section to a semiclassical phase-shift calculation; optimizing the MDP parameters for the scattering process yielded rather poor results, suggesting a limitation of the use of the KW MDPs for plasmas.

Internal Electron Tunneling Enabled Ultrasensitive Position/Force Peapod Sensors.

The electron quantum tunneling effect guarantees the ultrahigh spatial resolution of the scanning tunneling microscope (STM), but there have been no other significant applications of this effect after the invention of STM. Here we report the implementation of electron-tunneling-based high sensitivity transducers using a peapod B4C nanowire, where discrete Ni6Si2B nanorods are embedded in the nanowire in a peapod form. The deformation of the nanowire provides a higher order scaling effect between conductivity and deformation strain, thus allowing the potentials of position and force sensing at the picoscale.