⚛️ biophysics

Light Martini water accelerates sampling in coarse-grained molecular dynamics simulations

The authors introduce "light Martini water," a low-viscosity water model with reduced bead mass that significantly accelerates sampling rates in coarse-grained molecular dynamics simulations of intrinsically disordered proteins and lipid bilayers while preserving equilibrium properties and simulation accuracy.

Original authors: Elgendy, A., Zeipelt, A. P., Schäfer, L. V.

Published 2026-09-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Elgendy, A., Zeipelt, A. P., Schäfer, L. V.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

In the microscopic world of biology, life is a constant state of motion. Proteins, the workhorses of the cell, do not sit still; they twist, fold, and unfold in a complex dance of shapes that allows them to perform their functions. To understand how these molecules behave, scientists use a powerful tool called molecular dynamics. This is a computer simulation that acts like a high-speed movie camera, tracking the position of every single atom in a system over time. However, there is a significant problem with this approach: the movie is often too slow. Many biological processes, such as a protein changing its shape or a membrane rearranging itself, happen over time scales that are far longer than what current computers can easily simulate. Even with the most powerful supercomputers, watching these events unfold in real-time detail can take years of calculation time.

To get around this speed limit, researchers often use a simplified version of reality known as a coarse-grained model. Instead of tracking every single atom, they group clusters of atoms together and treat them as single, larger beads. This reduces the number of items the computer has to juggle, allowing the simulation to run much faster. One of the most popular frameworks for this is called Martini. While this method speeds things up, it still faces a bottleneck: the water surrounding these molecules. In these simulations, water is not just a passive background; it is a thick, viscous fluid that drags on the moving parts, slowing down their motion. The researchers in this study asked a simple but profound question: what if they could make this simulated water thinner and less resistant to motion, allowing the molecules to move more freely without changing the fundamental rules of how they interact?

The team set out to create a new version of water for their simulations, which they named "light Martini water." In the standard Martini model, a single water bead represents four real water molecules and has a specific weight. The researchers systematically reduced the mass of these water beads, essentially making them lighter and easier to push around. They tested this idea by running simulations to see if the computer could still calculate the movements accurately without the system falling apart or producing nonsense results. They found that they could safely reduce the mass of the water bead from 72 units down to 20 units. At this lower weight, the water behaved like a much thinner fluid, flowing more easily and offering less resistance to the molecules moving through it.

When they put this lighter water to the test with real biological systems, the results were striking. They simulated several intrinsically disordered proteins, which are flexible chains that do not have a fixed shape and are known to be particularly difficult to study because they are constantly changing form. In the standard, heavier water, these protein chains took a long time to rearrange themselves. In the light water, however, they moved much faster. Depending on the specific protein, the researchers observed that the sampling rate—the speed at which the protein explored its different shapes—increased by up to nearly three times. This means that a simulation that would have taken a month to complete could now be finished in a week, or even less, while still capturing the same physical behavior.

The researchers also tested this new water on a model of a cell membrane, which is a double layer of lipids that acts as a barrier for cells. They watched how fast the lipid molecules moved sideways within the layer. In the light water, these lipids diffused about 16 percent faster than in the standard water. This acceleration was significant, though not as dramatic as what they saw with the flexible proteins. This difference makes sense because the movement of lipids in a membrane is largely governed by how tightly they are packed against each other, rather than just how easily the water flows around them. In contrast, the flexible protein chains are more directly influenced by the friction of the surrounding water, so they benefited more from the change.

Crucially, the team verified that making the water lighter did not break the physics of the simulation. They checked the average shapes of the proteins and the structural properties of the membranes, such as their thickness and density. They found that these equilibrium properties remained exactly the same. The proteins did not become more or less compact, and the membranes did not change their structure. The only thing that changed was the speed at which the system explored those shapes. This is a vital distinction: the researchers did not alter the destination of the journey, only the speed of the vehicle. They proved that the lighter water model is a safe and effective way to accelerate the study of slow biological processes without compromising the accuracy of the results.

This work offers a practical solution to a persistent problem in computational biology. By simply adjusting the mass of the water beads, the researchers have provided a tool that can be used immediately in existing simulation software. It requires no complex new algorithms and adds no extra computational cost. For scientists studying complex, slow-moving systems like disordered proteins or the formation of cellular droplets, this method could dramatically shorten the time needed to gather meaningful data. The study suggests that while the "native" speed of the original model is altered, the fundamental behavior of the molecules remains intact, opening the door to observing biological events that were previously out of reach due to time constraints.

Technical Summary: Light Martini Water for Accelerated Coarse-Grained Sampling

Problem Statement
Molecular dynamics (MD) simulations of slow biomolecular processes, such as the conformational exploration of intrinsically disordered proteins (IDPs), are computationally demanding. While coarse-grained (CG) models like Martini reduce computational cost by mapping groups of atoms to interaction sites ("beads") and allowing larger integration time steps, sampling challenges persist for large systems and long time scales. The fundamental limitation is the sampling problem: biologically relevant processes often occur on microsecond to millisecond scales, yet simulations are typically restricted to shorter durations. Although techniques like hydrogen mass repartitioning (HMR) and mass scaling have been developed for all-atom models to accelerate sampling, there is a need for analogous, efficient solutions specifically for the Martini CG force field, which employs a single-bead water model with a mass of 72 amu (representing four water molecules).

Methodology
The authors developed "light Martini water," a low-viscosity water model based on the systematic reduction of the Martini water bead mass. The methodology involved:

Mass Scaling and Stability Testing: The mass of the Martini water bead ( $m_W$ ) was systematically reduced from the standard 72 amu down to 13 amu. The stability of the equations of motion was assessed in NVE (microcanonical) simulations using 20 fs time steps, which are standard for Martini. Metrics included total energy drift and fluctuations of total, kinetic, and potential energies.
Bulk Water Characterization: For the selected mass ( $m_W = 20$ amu), the authors calculated translational self-diffusion coefficients and viscosities in bulk water. They verified that equilibrium structural properties, specifically number density and radial distribution functions (RDFs), were preserved compared to the standard model.
Application to Biomolecular Systems: The light water model was applied to two distinct systems to evaluate sampling acceleration:
- Intrinsically Disordered Proteins (IDPs): Simulations were performed on a diverse set of IDPs (Hst5, Aβ40, A1LCD, Asyn, FUSLCD) with varying sequence lengths and charge properties. Both normal (72 amu) and light (20 amu) water models were used. Sampling efficiency was quantified by analyzing the decay of time correlation functions (TCFs) for the radius of gyration ( $R_g$ ) and end-to-end distance ( $R_{ee}$ ), as well as protein translational diffusion coefficients. Salt effects were also investigated by simulating specific IDPs with and without added salt.
- Lipid Bilayers: Simulations of POPC and DPPC/cholesterol bilayers were conducted to measure lateral lipid diffusion coefficients ( $D_L$ ), area per lipid (APL), and area compressibility modulus ( $K_A$ ).
Validation: Equilibrium properties were rigorously compared between the standard and light water models to ensure that the acceleration did not compromise structural accuracy.

Key Results

Integration Stability: Reducing the water bead mass to 20 amu maintained stable integration with 20 fs time steps. While energy drifts increased slightly below 20 amu, the fluctuations remained within acceptable bounds, and the total energy fluctuations were significantly smaller than kinetic or potential energy fluctuations.
Bulk Water Dynamics: The reduction of mass from 72 to 20 amu resulted in a 1.8-fold increase in the self-diffusion coefficient of water (from $2.23 \pm 0.03$ to $3.96 \pm 0.04$ nm $^2$ ns $^{-1}$ ) and a corresponding decrease in viscosity (from 0.66 to 0.36 mPa s). This speedup aligns with the theoretical expectation of $\sqrt{72/20} \approx 1.9$ .
IDP Sampling Acceleration: The use of light Martini water accelerated the chain reconfiguration dynamics of IDPs. Speedup factors for the decorrelation of $R_g$ and $R_{ee}$ ranged from 1.08 to 2.68, depending on the specific protein and the observable. For example, A1LCD showed a 2.68-fold speedup for $R_{ee}$ decay. Translational diffusion of IDPs increased by factors between 1.48 and 1.82. The degree of acceleration was found to depend on the balance between intramolecular interactions and solvent friction; proteins with higher charge fractions or specific salt conditions showed varying sensitivities to the reduced solvent viscosity.
Lipid Bilayer Acceleration: In POPC bilayers, lateral lipid diffusion increased by a factor of 1.16 (from $44.9 \pm 0.6$ to $51.9 \pm 0.6 \times 10^{-12}$ m $^2$ s $^{-1}$ ). In contrast, a more ordered DPPC/cholesterol bilayer showed negligible speedup (factor of 1.01), indicating that lipid diffusion in highly packed membranes is governed more by lipid-lipid friction than lipid-solvent friction.
Preservation of Equilibrium Properties: Crucially, the reduction in water mass did not alter equilibrium properties. Water densities, RDFs, IDP $R_g$ and $R_{ee}$ distributions, and membrane structural parameters (APL, $K_A$ , density profiles, and order parameters) remained statistically indistinguishable between the standard and light water models.

Significance and Claims
The paper claims that light Martini water is a robust, trivial-to-implement, and computationally free method to accelerate sampling in Martini CG simulations. By reducing the water bead mass to 20 amu, the authors achieved significant speedups in dynamic processes (up to 2.68-fold for IDP reconfiguration) without compromising the accuracy of equilibrium structural and thermodynamic properties.

The authors emphasize that while the "native" dynamics of the original Martini model are perturbed, this is generally acceptable for CG simulations where time scales require system-specific mapping to physical time. The method is presented as universally applicable to Martini 3 simulations, particularly beneficial for processes where slow diffusive dynamics represent a sampling bottleneck, such as IDP condensate formation, phase separation, and protein-ligand interactions. The paper modestly notes that the speedup is system-dependent and may be zero for systems dominated by internal friction (e.g., the DPPC/cholesterol bilayer), but asserts that the model will never slow down dynamics.

More like this