NMRlipids Databank makes data-driven analysis of biomembrane properties accessible for all

Abstract

Cellular membrane lipid composition is implicated in diseases and major biological functions in cells, but membranes are difficult to study experimentally due to their intrinsic disorder and complex phase behaviour. Molecular dynamics (MD) simulations have been useful in understanding membrane systems, but they require significant computational resources and often suffer inaccuracies in model parameters. Applications of data-driven and machine learning methods, currently revolutionising many fields, are limited to membrane systems due to the lack of suitable training sets. Here we present the NMRlipids Databank---a community-driven, open-for-all database featuring programmatic access to quality evaluated atom-resolution MD simulations of lipid bilayers. The NMRlipids databank will benefit scientists in different disciplines by providing automatic ranking of simulations based on their quality against experiments, flexible implementations of data-driven and machine learning applications, and rapid access to simulation data via graphical user interface. To demonstrate the unlocked possibilities beyond current MD simulation studies, we analyzed how anisotropic diffusion of water and cholesterol flip-flop rates depend on membrane properties.

Quantitative Comparison against Experiments Reveals Imperfections in Force Fields’ Descriptions of POPC–Cholesterol Interactions

“…Nevertheless, the minima in X-ray form factors are independent of the simulation box size and correlate with membrane dimensions. 48 To avoid the effects from simulation system sizes, we thus interpolated the locations of the first two minima in the form factors over the entire studied range of CHOL concentrations (bottom right panel of Figure 2) for a more direct comparison between simulations and experiments. These curves highlight that at first the addition of CHOL shifts the first minima to smaller wave vector values in the experiment; this is reasonably well captured by the simulation force fields, although CHARMM36 seems to be off more than the other three force fields.…”

“…82−85 Our results demonstrate that quality evaluation of lipid mixture simulations against experimental NMR and X-ray scattering data gives consistent results for how accurately force field parameters capture intermolecular interactions. The interpolation approach introduced here extends the NMRlipids Databank quality metrics 48 beyond individual systems: this enables the automatic ranking of not only lipid mixtures but also membranes mixed with other molecules such as ions. Such tools will strongly support the emerging endeavors for automatic improvement of force field parameters.…”

“…Structural Properties. The C−H bond order parameters, form factors, and electron density profiles, automatically calculated by the NMRlipids Databank, 48 were used. Similarly to the experimental X-ray scattering data, the membrane thickness was defined as twice the distance from the electron density maximum to the membrane center.…”

“…38−42 This approach has been useful for finding the best force fields to describe the headgroups of phosphatidylcholine (PC), 38,43 phosphatidylserine (PS), 41 phoshatidylethanolamine (PE), 42 and phosphatidylglycerol (PG) lipids; 42 to evaluate and improve membrane interactions with ions [40][41][42]44,45 and small molecules; 46 and to find simulation parameters predicting most realistic packing properties of membranes. 47,48 Furthermore, quantitative quality measures based on the C−H bond order parameters have been recently defined and used to rank simulations in the NMRlipids Databank. 48 However, such automatic quality evaluation is limited to simulations for which experimental data at the corresponding composition and temperature are available.…”

“…47,48 Furthermore, quantitative quality measures based on the C−H bond order parameters have been recently defined and used to rank simulations in the NMRlipids Databank. 48 However, such automatic quality evaluation is limited to simulations for which experimental data at the corresponding composition and temperature are available. Because simulations mimicking all experimental compositions for multicomponent membranes are often tedious to produce, quality evaluation of mixed lipid bilayers…”