Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2026 May 26;22(11):5858–5866. doi: 10.1021/acs.jctc.6c00463

Martini 3 Metabolome

Christopher Brasnett 1, Chelsea M Brown 1, Linus Grünewald 1, Jan A Stevens 1, Siewert J Marrink 1,*
PMCID: PMC13256178  PMID: 42191118

Abstract

Metabolites are ubiquitous in all living cells and are essential mediators of biochemical processes, serving either as substrates or as cofactors to enable reactions. Capturing this diversity in computational workflows is important for allowing realistic simulations of cytoplasm. Coarse-grained molecular dynamics enables the simulation of large-scale systems up to the level of whole cells but is limited by the availability of refined parameters for all possible components in the system. In this work, we describe the parametrization of 186 common metabolites found in bacteria and eukaryotes within the framework of the Martini 3 force field. To showcase the behavior of Martini metabolites in a biological setting, we report simulations of protein–ligand binding and membrane permeation. The establishment of a Martini metabolome enables high-throughput simulations of metabolites interacting with other biomolecules and opens the way for simulations of realistic cellular environments.

graphic file with name ct6c00463_0007.jpg

graphic file with name ct6c00463_0005.jpg

Introduction

Molecular dynamics (MD) simulations have emerged as a powerful computational approach to investigate biological systems with spatiotemporal resolution nearly impossible to achieve in traditional laboratory settings. These simulations can provide detailed insights into how the full complexity of cellular environments influences molecular behavior, serving as a computational microscope for biological processes. Coarse-grained (CG) models offer access to larger systems and time scales by simplifying the representation. One of the most popular CG force fields is Martini, which broadly uses a 4-to-1 heavy atom to particle mapping, sufficient to retain chemical specificity. While the latest version of the force field (Martini 3) has good coverage for simulating biomolecular systems, from lipids, proteins, and sugars to small organic molecules, extensive parameters for metabolites, such as cofactors and nucleotides, are still missing. Some metabolites have been modeled in previous iterations of the force field, − and there is a small selection compatible with the current version, ,− but the vast number of metabolites found in vivo are not yet captured in the Martini framework.

Simulations are increasingly moving toward biological levels of accuracy in terms of their components, − attempting to reach an in situ composition. One of the major bottlenecks in assembling systems such as these is the availability of parameters for all of the components in the system. This challenge is particularly evident for metabolites, where the vast chemical diversity of molecules requires extensive parametrization efforts. The magnitude of this problem becomes especially clear when modeling whole cells like the JCVI-Syn3A minimal cell. Despite the reduced complexity of this synthetic organism, its metabolome contains around 200 distinct molecules. , Equally, this is a problem for modeling organelles such as the mitochondria that have over 100 metabolites reported in the matrix alone. ,, The majority of the metabolites from both systems are not yet available in the standard Martini 3 database.

Here, we present refined Martini 3 parameters for 186 metabolites found in bacterial and eukaryotic cells. , They include updated nucleotide cofactors, flavins, and phosphorylated sugars, cofactors that are crucial for biological functions. Following the Martini framework, ,, we used all-atom MD simulations as references to generate initial parameters that were subsequently refined based on numerical stability and on properties such as solvent-accessible surface area (SASA) and LogP measurements. The resulting Martini metabolome will allow high-throughput, smaller-scale studies investigating individual metabolite interactions, alongside enabling large-scale simulations closer to in vivo conditions. We demonstrate the versatility of applications for this set of molecules with two showcases: the binding of ATP to an ABC transporter and the permeation of glycerol across a lipid membrane.

Results

Selection of Metabolites

To ensure a large coverage of biologically relevant metabolites, those found in a bacterial cell (JCVI-Syn3A) and in the human mitochondrial matrix were chosen for the Martini metabolome. The resulting 186 metabolites represent a broad range of chemical classes, including nucleotides, ions, carbohydrates, amino acid derivatives, lipids, amino acids, and cofactors (Figure and Table S1). Combined with the already existing models, including molecular oxygen and those related to lipid metabolism, the Martini 3 metabolome now covers 99.3% of the metabolites present in JCVI-Syn3A and 98.7% of the mitochondrial matrix. Even though the selection was distinct from the database, the metabolite library covers the vast majority of the abundant metabolites measured in E. coli.

1.

1

Open in a new tab

Overview of the parametrized metabolites. (Left) Distribution by number of molecules parametrized in this study (186 in total), showing the diversity of newly available metabolites. (Middle) Molar concentration distribution in the JCVI-Syn3A cell cytoplasm. (Right) Molar concentration distribution in the human mitochondrial matrix. Metabolites are classified into functional categories (nucleotides, amino acids, amino acid derivatives, carbohydrates, TCA cycle intermediates, cofactors, lipids, ions, and other metabolites) and colored accordingly. In the middle and right panels, the hashed sections of segments represent molar concentrations in the respective metabolomes with models not yet available in Martini 3. Percentages are shown for classes representing ≥5% of the total.

The ionization state for each metabolite was assigned using SwissParam , to identify the dominant species at pH 7. Alternative protonation states are also provided for some nucleotides (see Table S1) to capture species that might be desired by the simulation community. Protonation states in simulations may be highly context-dependent, and so model selection should be made carefully with reference to Table S1 and the underlying atomistic topologies.

Parameterization Strategy

The parametrization strategy adhered to the general Martini guidelines. Following an atomistic simulation, a molecule was mapped to a pseudo-CG trajectory (Figure A). CG molecular topologies were designed for each molecule to reproduce the respective bonded parameters of its pseudo-CG trajectory. In Figure B, we show an example of optimized bonded interactions for adenine-containing nucleotides, representative of the level of optimization possible across the metabolome. Bead assignments were made by following standard mapping procedures and keeping chemically functional groups together. The bead type was initially assigned based on the partitioning behavior of the underlying group. Thus, throughout the validation of nonbonded parameters, the Martini bead type for every molecule was assigned using the validated and suggested bead types for the force field. ,, Solvent-accessible surface areas (SASAs) were assessed against all-atom references to ensure good volumetric matching (Figure C,D). In addition to the usual parametrization approach, we stress tested every molecule at high concentrations to further ensure numerical stability in crowded environments. The parametrization protocol is described in full in the Methods section. Topologies and other parametrization reference data for all of the 186 metabolites parametrized in this study can be found in the GitHub repository https://github.com/Martini-Force-Field-Initiative/M3-Metabolome.

2.

2

Open in a new tab

Parameterization strategy of metabolites. (A, B) Examples of molecular mapping and bonded interaction matching for adenosine. (C) Connolly surfaces of adenosine at atomistic (blue) and CG (red) resolutions. (D) SASA for all metabolites, colored by the Hellinger distance between the atomistic and CG distributions. The dashed and dotted lines represent 5% and 10% deviations, respectively. (E) Water-octanol partitioning free energy of transfer for metabolites, colored by net charge. The mean absolute error and Spearman rank are indicated by MAE and ρ, respectively.

In general, the molecules described in this work fall into two categories: non-nucleotides and nucleotides. The former class was generally parametrized ab initio, using the process described above. The latter class (and other associated molecules such as sugar-based metabolites) was parametrized using existing fragments in a self-consistent and hierarchical manner. For example, parameters for free nucleobases and sugars have already been developed for the Martini 3 force field. , However, parameters for neither nucleosides nor nucleotides have been systematically developed. While the model of adenosine shown in Figure A uses the existing mappings of adenine and ribose, the combination shown here is new. Where available, common interactions were grouped together across molecules to ensure consistency across the set. For instance, the ADE2-ADE1-RIB1-RIB2 dihedral present in adenosine (Figure B) is also present in AMP, ADP, and ATP. Such interactions were therefore fitted once, against an average distribution, and optimized across all molecules where present. Bonded interaction validation figures are available for every molecule in the associated repository.

Validation against Partitioning Data

One of the major targets of Martini parametrization is the partitioning behavior of molecules between water and other organic solvents. To this end, we computed the free energy of transfer between water and octanol (log P) via alchemical transformations. The results are listed in Table S1. Ideally, it would be possible to compare the calculated partitioning free energies to an experimental reference, but these data are only available for a limited set of molecules. In Figure S1, we show that for molecules with experimental reference data, the calculated partitioning behavior of the models matches closely, at least for neutral molecules.

To generate a comprehensive comparison data set, we additionally used the SwissADME server to generate predicted values for all the molecules presented in this work using the XLOGP3 method. Figure E shows a comparison between the estimated partition coefficients of all the molecules and the predictions made by the server. Compared with the prediction, the transfer free energy of the metabolites has a mean absolute error (MAE) of 9.7 kJ/mol. While this does not represent a high degree of accuracy with respect to the predicted values, it is likely that the predicted values for many of these molecules (particularly those that are either polar or charged) do not represent an accurate prediction themselves.

In an effort to better understand the limitations of the predictors, we compared XLOGP3 predictions to experimental LogP values used in recent Martini 3 parametrization papers. ,,,, As shown in Table S1 and Figure S2, XLOGP3 predictions are in good agreement with small-molecule partitioning data. Additionally, LogP values for simple sugars as well as for a more complex molecule like riboflavin are close to the experimental values. However, as we note in Figure E, predictions start deviating strongly for charged molecules. Further, for some sugars, LogP values are predicted accurately, while for other sugars, they are not. Other LogP predictors generated by SwissADME do not lead to an overall improvement (Figure S3). A final comparison of neutral and zwitterionic molecules for the XLOGP3 predictor further demonstrates the effect of the electrical charge on the prediction deviation (Figure S4). The MAE for neutral molecules (4.3 kJ/mol) is within an acceptable range for Martini molecules. Given that XLOGP3 predictions for neutral molecules are quite reliable, the bead assignment scheme followed in this work generated an overall good agreement with the predictor.

We therefore conclude that, in general, LogP predictors like XLOGP3 can be a good starting point if no LogP data are available. However, with increasing molecule complexity, predictions can be unreliable, while for charged molecules, a strong deviation is to be expected. In assigning CG bead types, we have remained faithful to the underlying chemical nature of each molecule following standard bead assignment rules and recommendations. In the absence of systematic experimental data, we therefore believe that the current estimates of partitioning behavior for the Martini metabolome are sufficiently faithful for each molecule.

Showcases

ATP Binding

The Martini 3 metabolome opens the way for studying the interaction of metabolites with other biomolecules. To showcase these opportunities, we selected two examples of basic cellular importance, namely, the protein binding of ATP and the membrane permeation of glycerol.

Due to its biological importance, the metabolite chosen to showcase protein–ligand interactions was ATP (Figure A). ATP was simulated with a protein from the ATP-binding cassette (ABC) transporter family, namely, NosDFY, which is a member of the nitrous oxide reductase pathway. The structure has been solved in an ATP-bound state, with the ligand bound to both copies of the nucleotide-binding domain (PDB ID: 7O17), which was used to build the CG starting state in our simulations. Note that an elastic network was used to retain the secondary structure and the tertiary fold of the protein. Three different systems were simulated: the whole transporter in a lipid bilayer bound to ATP, the soluble NosF dimer with ATP bound, and the soluble dimer with ATP started in bulk solution (Figure S5). When the holo form of the whole complex was simulated, ATP was observed to stay within the binding site throughout five replicate simulations of 1 μs (Figure S6A).

3.

3

Open in a new tab

Metabolite interaction with proteins. (A) Mapping of ATP from atomistic to CG resolution. The bead names are indicated by each bead, with the Martini bead type indicated in brackets. The virtual interaction site RIB4 is indicated by an orange dot. (B) The surface of the protein shows the ATP occupancy with NosF throughout simulations, with the darker color showing increased occupancy. The stick representation of ATP denotes the canonical binding sites from the cryo-EM structure. (C) A plot showing the root mean squared deviation (RMSD) of each ATP (one with a solid line, the other with a dotted line) from the initial starting position in replica simulations. The highlighted red line shows an ATP molecule briefly leaving the binding site before returning again.

To study possible entry and exit pathways, additional simulations were performed on just the soluble ATP-binding protein homodimer (NosF), again with ATP bound initially. Within the majority of simulation time, the ligand stays bound with occupancy of the binding site at ∼60% (Figure B). Interestingly, in one replica, an ATP leaves the known binding site, and the molecule returns to the starting position after sampling the solvent (Figure C), showing that our ATP model captures the correct biological behavior. To further test the ability of the molecule to find the binding site from the bulk, two copies of the ligand were placed in the solvent. Over a cumulative 5 μs simulation time, ATP was observed to spontaneously bind to the canonical binding site (Figure S6B).

Glycerol Permeation of Lipid Membranes

The permeation of metabolites across lipid bilayers provides another test of parameter accuracy under biologically relevant conditions. Here, we investigated glycerol transport across the bacterial membrane of the JCVI-Syn3A minimal cell. Glycerol represents a particularly relevant case study for the minimal cell, where this essential metabolite must rely on passive diffusion due to the absence of a dedicated transporter in the reduced genome.

We constructed a membrane model of the native Syn3A lipid composition containing cholesterol, sphingomyelin, cardiolipin, palmitoyl-oleoyl-phosphatidylcholine (POPC), and dioleoyl-phosphatidylglycerol (DOPG) at experimentally determined concentrations (Figure b). The potential of mean force (PMF) for glycerol translocation was calculated using umbrella sampling, and permeation rates were determined by combining the free energy profile with position-dependent diffusion coefficients calculated within each umbrella window (Figure c). Our simulations yielded a glycerol permeation rate of 7 ± 2 nm/s through the Syn3A membrane, in agreement with a recent in vitro measurement of 3 ± 3 nm/s across a membrane that closely matches the lipid composition. Nevertheless, the generic speed-up of dynamics inherent to CG models is known to vary between systems, introducing additional uncertainty beyond what is captured by our error analysis. The result should therefore be interpreted as a qualitative validation of the parametrization rather than a strict quantitative prediction.

4.

4

Open in a new tab

Glycerol permeation across a bacterial membrane. (A) Mapping of glycerol from atomistic to CG resolution. The bead names are indicated next to each bead with the Martini bead type indicated in brackets. (B) Snapshot from umbrella sampling simulation showing glycerol (red) above the membrane containing cholesterol (white), cardiolipin (pink), POPC (orange), DOPG (cyan), and sphingomyelin (yellow). In red, a representative permeation pathway through the membrane is shown. (C) The calculated potential of mean force and diffusion profile for glycerol permeating through the JCVI-Syn3A membrane.

Discussion

The metabolite library presented in this work addresses a fundamental limitation that has constrained the biological realism in molecular simulations targeting cellular environments with the Martini 3 force field. Without validated parameters spanning the chemical diversity of biological systems, researchers must either omit metabolites that are present in their experimental target systems or undertake laborious parametrization efforts that become prohibitive when attempting to capture realistic systems with high molecular diversity. While ongoing initiatives to centralize and distribute Martini parameters have improved accessibility, a significant lack of metabolite parameters has persisted. By providing validated Martini 3 parameters for 186 cellular metabolites representing major chemical classes in bacterial and eukaryotic cells, this work opens the door to modeling in situ cellular environments that have previously been limited in complexity and realism. Current efforts are already underway to simulate the JCVI-Syn3A cytosol using the Martini 3 metabolome.

While we have demonstrated the potential of the CG models in this work to already reproduce basic biophysical phenomena, we believe that the models still have significant potential for further improvement. Beyond numerical stability and basic physicochemical properties, we did not perform extensive validation testing for any molecule in the database. Depending on the intended application, more specific checks and refinements of the current parameters may be necessary. Future use of more targeted experimental reference data, along with additional considerations during parametrization, could almost certainly result in more optimized models for specific applications.

Although our current library covers the major metabolite classes found in bacterial and eukaryotic cells, the framework developed here provides a foundation for filling in the remaining gaps. Furthermore, this metabolite library has the potential to serve as a comprehensive data set for the training and validation of machine learning approaches to automate CG topology generation. , Machine learning approaches for force field parametrization require diverse training data that span the chemical space of biological molecules. This metabolite library provides such a foundation with validated Martini parameters across multiple chemical classes that represent the molecular diversity found in cellular environments.

The availability of these validated parameters transforms cellular metabolites from a technical barrier to a standard component of Martini modeling workflows. These parameters allow researchers to focus on fundamental biological questions, from metabolite–protein interactions to cellular crowding effects. As simulations increasingly approach biological complexity, this metabolite data set provides essential components that bridge the gap between simplified model systems and the rich chemical environments found in cells.

Methods

Parameterization Strategy

Parametrization followed the standard Martini 3 parametrization pipeline. First, atomistic simulations of each individual molecule were performed. The trajectories were then mapped according to a consensus mapping scheme at a coarse-grained resolution. The mapped trajectories were then analyzed, such that CG molecular topologies could be generated. Once bonded distributions between resolutions were matched as optimally as possible, the molecules were further validated by comparisons of solvent-accessible surface area (SASA) and the transfer free energy between octanol and water. Molecular topologies were further stress-tested to ensure stability in crowded environments.

Atomistic Reference Simulations

SMILES strings were compiled and uploaded to the SwissParam server ,, to obtain simulation input parameters and structures for the CHARMM36 force field. Simulations were conducted using GROMACS 2024.1. Single copies of molecules were placed in dodecahedral simulation boxes, 5 nm from the edge of the box, using the gromacs editconf program. The molecule was subsequently solvated with an ionic concentration of 0.15 M NaCl, plus additional ions for system neutralization. The system was subsequently subject to short equilibration simulations in the NVT and NPT ensembles for 1 ns, before a 1 μs production simulation. Solvent-free trajectories are available as part of the Supporting Information. For all simulations, the time step was 2 fs. Where pressure coupling was used, the pressure was set to 1.0 bar with a compressibility of 4.5 × 10–5 bar–1. Electrostatic interactions were treated with PME with a cutoff of 1.2 nm. During equilibration, the C-rescale barostat was used, and production simulations were performed using the Parrinello–Rahman barostat with isotropic pressure coupling. , Simulation temperature was controlled with the V-rescale thermostat.

Coarse-Grained Simulations

Unless stated otherwise, all CG simulations in this work followed the standard Martini 3 protocol described below. Simulations were conducted using GROMACS 2024.1. Simulations of single molecules were prepared directly from the solvent-free mapped atomistic systems, using the same simulation cell dimension. Molecules were then solvated and charge neutralized with 0.15 M NaCl. The system subsequently underwent energy minimization using Gromacs’ steepest descent algorithm, followed by equilibration and production simulations. Electrostatics were treated with a reaction field with a cutoff of 1.1 nm, and van der Waals interactions were treated with potential-shift-verlet, also with a cutoff of 1.1 nm. Additionally, we used the recent recommended settings from Kim et al. (verlet-buffer-tolerance = −1, rlist = 1.35 nm) to correct neighborlist artifacts. Pressure was maintained isotropically at 1 bar using the c-rescale barostat during equilibration, and for production, the Parrinello–Rahman barostat was used. Compressibility was set to 3.0 × 10–4 bar–1. The V-rescale thermostat was used to maintain the temperature at 300 K for all simulations. A time step of 10 fs was used for equilibration, which was increased to 20 fs for production runs.

Mapping

Molecular mapping was conducted according to standard Martini 3 mapping rules. The mapping of bonded distributions was performed using the fast forward Python package and, where necessary, collated by custom scripts. Mapping (.map) files following the Vermouth format were generated using the cgbuilder tool and are supplied in the Supporting Information. For future cheminformatic purposes, we also provide the CGsmiles strings for mapping information.

Bonded Interactions

Initial topologies of molecules were generated using the ff_inter program of the fast forward Python package. , As fast_forward is designed to generate high-fidelity bonded parameters based on mapped simulations, in the first instance, these parameters were tested without further inspection. Initial conformers of molecules were generated using Polyply and solvated in a 5 × 5 × 5 nm3 simulation cell with 0.15 M NaCl and neutralizing counterions. The systems were then subjected to steepest descent energy minimization, followed by equilibration and production simulations of lengths 10 and 500 ns, respectively.

Following an initial simulation of a molecule, the resultant bonded distributions were validated against their mapped references using the ff_assess program of FastForward. ff_assess scores simulated bonded distributions against the references used during fitting and returns an average Hellinger distance between the two probability distributions. A score below 0.3 is considered a good agreement with an atomistic reference for a molecule, which was used as a target throughout parametrization. Molecular topologies were then optimized according to which bonded parameters were in least agreement with their respective references. This was achieved by several methods, such as varying force constants or adding additional exclusions between beads. In the case of dihedral terms, initial fits using ff_inter may have overfit the mapped distribution and caused instabilities. In this case, the ff_inter program was rerun with a smaller value for the maximum multiplicity of dihedral to be used during fitting to achieve better stability at the sacrifice of a degree of fidelity.

Stress Testing

To ensure CG model stability, once the bonded parameters of a single copy of a molecule had been validated, 50 copies were simulated in a 10 × 10 × 10 nm cubic box for 500 ns. If the simulation proceeded without error, then the model was determined to be stable for the release. If the simulation did not run to completion, then the model was reviewed for stability by repeating the mapping and interaction optimization steps described above as necessary. The simulation input parameters followed the standard Martini parameters described above.

Nonbonded Validation

Partition coefficients (LogP) for each molecule were calculated using alchemical free energy transformations. The free energy of solvation was estimated in both water and hydrated (0.3 M water) octanol, using 21 steps during which the nonbonded interactions between the solute and solvent were turned off. Soft core potentials were used using the recommended (sc-alpha = 0.5, sc-power = 1) values, along with a stochastic dynamics integrator and an inverse friction constant of 1.0 ps. The transfer free energy (ΔG) was calculated using MBAR as implemented in the alchemlyb Python package.

Showcase 1: Ligand Binding

The atomistic structure file for NosDFY was downloaded from the PDB database. The protein was converted to a Martini representation using martinize2 with an elastic network (merging all chains when relevant) using a force constant of 700 kJ/(mol nm2). The elastic network was applied in the range of 0.5–1 nm. When the metabolite was started within the binding site, the coordinates were obtained by mapping with CGBuilder. When the ligand was started outside of the binding site, the gmx insert-molecules tool was used to generate independent starting positions for two molecules. The whole NosDFY transporter was embedded in a membrane (POPE:POPG, 4:1) using insane. Systems were solvated and 0.15 M NaCl was added, with neutralizing counterions. The systems were energy-minimized, equilibrated, and production simulations performed using the standard Martini protocol described above. For the simulations without a bilayer, isotropic pressure coupling was used at 1 bar. For the entire protein complex in a bilayer, semi-isotropic pressure coupling was used at the same reference pressure. All simulations were performed at 310 K. Each system was simulated for 5 × 1 μs.

Occupancy analysis was performed using PyLipID and visualized using VMD. The RMSD of the ligand was calculated using gmx rmsd relative to the protein backbone and plotted using Matplotlib. Visualization of protein structures was aided by the MartiniGlass package.

Showcase 2: Membrane Permeability

A membrane model was constructed to reflect the native JCVI-Syn3A envelope composition: cholesterol (59%), sphingomyelin (18%), cardiolipin (17%), linked POPC (4%), and phosphatidylglycerol (2%). The bilayer was assembled using the insane tool in a 10 × 10 × 20 nm3 simulation box. The system was then solvated with Martini water, where a subset of water beads was replaced with ions to neutralize the model and add an ionic strength of 0.15 M NaCl.

The system was equilibrated and simulated using the standard Martini 3 simulation protocol as described previously, with the addition of a semi-isotropic pressure coupling. The bilayer was equilibrated for 200 ns at 300 K. After equilibration, umbrella sampling was performed with glycerol restrained at intervals of 0.1 nm from the bilayer center until 5 nm into the aqueous phase. Harmonic restraints with a force constant of 1000 kJ mol–1 were applied to the glycerol center of mass. Each umbrella window was simulated for 100 ns. The potential of mean force was reconstructed using WHAM with uncertainties estimated from 500 bootstrap iterations (Figure S7A).

Position-dependent diffusion coefficients along the transmembrane axis were calculated using the method described by Hummer. , For each umbrella window, five replicas of 10 ns were performed with the glycerol center of mass saved at every time step. These trajectories were used to calculate the local diffusion coefficient according to eq :

D(z=⟨z⟩)=var(z)2∫0∞Czz(t)dt 1

where C zz (t) is the autocorrelation function of the reaction coordinate (Figure S7B). Because of the high uncertainty in the calculated diffusion coefficients, a moving average filter with a window size of 2 was applied to the mean diffusion profile.

The membrane permeability was then calculated by integrating the position-dependent resistance across the membrane using the relation in eq :

1P=∫exp[W(z)kBT]D(z)dz 2

where W(z) is the potential of mean force, k B is the Boltzmann constant, and T is the temperature. To propagate uncertainties into the permeability estimate, this calculation was performed 100,000 times using randomly sampled combinations of the replicate diffusion profiles and bootstrapped PMF profiles (Figure S7C). This bootstrap analysis resulted in a permeability coefficient of 7 ± 2 nm/s.

Supplementary Material

ct6c00463_si_001.csv (104.2KB, csv)
ct6c00463_si_002.pdf (887.9KB, pdf)

Acknowledgments

S.J.M. acknowledges funding from NWO through the Summit Grant “Evolving life from non-life (EVOLF)”. J.A.S. and S.J.M. acknowledge funding from NWO through the NWA grant “The limits to growth: The challenge to dissipate energy”. C.M.B., L.G., and S.J.M. acknowledge funding from the ERC with the Advanced grant “COMP-O-CELL”. C.B. and S.J.M. are supported by the Novo Nordisk Foundation Grant NNF20OC0063808, “BOUNDLESS”. We would like to thank SURF for access to Snellius computing resources. We thank Fabian Grünewald for assistance with preparing CGsmiles strings for molecular mappings.

Full Supporting Information is available on Zenodo. A subset of Supporting Information (CG topology and coordinate files, mapping files, and bonded distribution comparisons) is available on the M3-Metabolome repository on GitHub: https://github.com/Martini-Force-Field-Initiative/M3-Metabolome. Parameters will also be made available on the MArtini Database (MAD) web server. ,

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jctc.6c00463.

  • Table S1, containing a full list of molecules parametrized along with additional details (LogP values, SMILES strings used, CGSmiles descriptors, notes on protonation states) (CSV)

  • Figures for additional LogP validation (Figures S1–S3), details of ATP-binding simulation setups (Figures S4 and S5), and additional validation for glycerol permeation (Figure S6) (PDF)

†.

C.B., C.M.B., L.G., and J.A.S. contributed equally to this work.

The authors declare no competing financial interest.

References

  1. Souza P. C. T., Alessandri R., Barnoud J., Thallmair S., Faustino I., Grünewald F., Patmanidis I., Abdizadeh H., Bruininks B. M. H., Wassenaar T. A., Kroon P. C., Melcr J., Nieto V., Corradi V., Khan H. M., Domański J., Javanainen M., Martinez-Seara H., Reuter N., Best R. B., Vattulainen I., Monticelli L., Periole X., Tieleman D. P., de Vries A. H., Marrink S. J.. Martini 3: A General Purpose Force Field for Coarse-Grained Molecular Dynamics. Nat. Methods. 2021;18(4):382–388. doi: 10.1038/s41592-021-01098-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Sousa F. M., Lima L. M. P., Arnarez C., Pereira M. M., Melo M. N.. Coarse-Grained Parameterization of Nucleotide Cofactors and Metabolites: Protonation Constants, Partition Coefficients, and Model Topologies. J. Chem. Inf. Model. 2021;61(1):335–346. doi: 10.1021/acs.jcim.0c01077. [DOI] [PubMed] [Google Scholar]
  3. de Jong D. H., Liguori N., van den Berg T., Arnarez C., Periole X., Marrink S. J.. Atomistic and Coarse Grain Topologies for the Cofactors Associated with the Photosystem II Core Complex. J. Phys. Chem. B. 2015;119(25):7791–7803. doi: 10.1021/acs.jpcb.5b00809. [DOI] [PubMed] [Google Scholar]
  4. Thallmair S., Vainikka P. A., Marrink S. J.. Lipid Fingerprints and Cofactor Dynamics of Light-Harvesting Complex II in Different Membranes. Biophys. J. 2019;116(8):1446–1455. doi: 10.1016/j.bpj.2019.03.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Grünewald F., Punt M. H., Jefferys E. E., Vainikka P. A., König M., Virtanen V., Meyer T. A., Pezeshkian W., Gormley A. J., Karonen M., Sansom M. S. P., Souza P. C. T., Marrink S. J.. Martini 3 Coarse-Grained Force Field for Carbohydrates. J. Chem. Theory Comput. 2022;18(12):7555–7569. doi: 10.1021/acs.jctc.2c00757. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Brandner A. F., Smith I. P. S., Marrink S. J., Souza P. C. T., Khalid S.. Systematic Approach to Parametrization of Disaccharides for the Martini 3 Coarse-Grained Force Field. J. Chem. Inf. Model. 2025;65(3):1537–1548. doi: 10.1021/acs.jcim.4c01874. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Chiariello M. G., Zarmiento-Garcia R., Marrink S.-J.. Martini 3 Coarse-Grained Model for the Cofactors Involved in Photosynthesis. Int. J. Mol. Sci. 2024;25(14):7947. doi: 10.3390/ijms25147947. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Waclawiková B., Cesar Telles de Souza P., Schwalbe M., Neochoritis C. G., Hoornenborg W., Nelemans S. A., Marrink S. J., El Aidy S.. Potential Binding Modes of the Gut Bacterial Metabolite, 5-Hydroxyindole, to the Intestinal L-Type Calcium Channels and Its Impact on the Microbiota in Rats. Gut Microbes. 2023;15(1):2154544. doi: 10.1080/19490976.2022.2154544. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Weerakoon D., Marzinek J. K., Pedebos C., Bond P. J., Khalid S.. Polymyxin B1 in the Escherichia Coli Inner Membrane: A Complex Story of Protein and Lipopolysaccharide-Mediated Insertion. J. Biol. Chem. 2024;300(10):107754. doi: 10.1016/j.jbc.2024.107754. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Stevens J. A., Grünewald F., Van Tilburg P. A. M., König M., Gilbert B. R., Brier T. A., Thornburg Z. R., Luthey-Schulten Z., Marrink S. J.. Molecular Dynamics Simulation of an Entire Cell. Front. Chem. 2023;11:1106495. doi: 10.3389/fchem.2023.1106495. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Brown C. M., Westendorp M. S. S., Zarmiento-Garcia R., Stevens J. A., Bruininks B. M. H., Rouse S. L., Marrink S. J., Wassenaar T. A.. An Integrative Modelling Approach to the Mitochondrial Cristae. Commun. Biol. 2025;8(1):972. doi: 10.1038/s42003-025-08381-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Dommer A., Casalino L., Kearns F., Rosenfeld M., Wauer N., Ahn S.-H., Russo J., Oliveira S., Morris C., Bogetti A., Trifan A., Brace A., Sztain T., Clyde A., Ma H., Chennubhotla C., Lee H., Turilli M., Khalid S., Tamayo-Mendoza T., Welborn M., Christensen A., Smith D. G., Qiao Z., Sirumalla S. K., O’Connor M., Manby F., Anandkumar A., Hardy D., Phillips J., Stern A., Romero J., Clark D., Dorrell M., Maiden T., Huang L., McCalpin J., Woods C., Gray A., Williams M., Barker B., Rajapaksha H., Pitts R., Gibbs T., Stone J., Zuckerman D. M., Mulholland A. J., Miller T., Jha S., Ramanathan A., Chong L., Amaro R. E.. #COVIDisAirborne: AI-Enabled Multiscale Computational Microscopy of Delta SARS-CoV-2 in a Respiratory Aerosol. Int. J. High-Perform. Comput. Appl. 2023;37(1):28–44. doi: 10.1177/10943420221128233. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Brown C. M., Marrink S. J.. Modeling Membranes in Situ . Curr. Opin. Struct. Biol. 2024;87:102837. doi: 10.1016/j.sbi.2024.102837. [DOI] [PubMed] [Google Scholar]
  14. Breuer M., Earnest T. M., Merryman C., Wise K. S., Sun L., Lynott M. R., Hutchison C. A., Smith H. O., Lapek J. D., Gonzalez D. J., de Crécy-Lagard V., Haas D., Hanson A. D., Labhsetwar P., Glass J. I., Luthey-Schulten Z.. Essential Metabolism for a Minimal Cell. eLife. 2019;8:e36842. doi: 10.7554/eLife.36842. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Thornburg Z. R., Bianchi D. M., Brier T. A., Gilbert B. R., Earnest T. M., Melo M. C. R., Safronova N., Sáenz J. P., Cook A. T., Wise K. S., Hutchison C. A., Smith H. O., Glass J. I., Luthey-Schulten Z.. Fundamental Behaviors Emerge from Simulations of a Living Minimal Cell. Cell. 2022;185(2):345–360. doi: 10.1016/j.cell.2021.12.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Chen W. W., Freinkman E., Wang T., Birsoy K., Sabatini D. M.. Absolute Quantification of Matrix Metabolites Reveals the Dynamics of Mitochondrial Metabolism. Cell. 2016;166(5):1324–1337. doi: 10.1016/j.cell.2016.07.040. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Westendorp M. S. S., Stevens J. A., Brown C. M., Dommer A. C., Wassenaar T. A., Bruininks B. M. H., Marrink S. J.. Compartment-Guided Assembly of Large-Scale Molecular Models with Bentopy. Protein Sci. 2026;35(3):e70480. doi: 10.1002/pro.70480. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Bennett B. D., Kimball E. H., Gao M., Osterhout R., Van Dien S. J., Rabinowitz J. D.. Absolute Metabolite Concentrations and Implied Enzyme Active Site Occupancy in Escherichia coli . Nat. Chem. Biol. 2009;5(8):593–599. doi: 10.1038/nchembio.186. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Alessandri R., Barnoud J., Gertsen A. S., Patmanidis I., De Vries A. H., Souza P. C. T., Marrink S. J.. Martini 3 Coarse-Grained Force Field: Small Molecules. Adv. Theory Simul. 2022;5(1):2100391. doi: 10.1002/adts.202100391. [DOI] [Google Scholar]
  20. Davoudi S., Vainikka P. A., Marrink S. J., Ghysels A.. Validation of a Coarse-Grained Martini 3 Model for Molecular Oxygen. J. Chem. Theory Comput. 2025;21(1):428–439. doi: 10.1021/acs.jctc.4c01348. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Pedersen K. B., Ingólfsson H. I., Ramirez-Echemendia D. P., Borges-Araújo L., Andreasen M. D., Empereur-mot C., Melcr J., Ozturk T. N., Bennett W. F. D., Kjølbye L. R., Brasnett C., Corradi V., Khan H. M., Cino E. A., Crowley J., Kim H., Fábián B., Borges-Araújo A. C., Pavan G. M., Launay G., Lolicato F., Wassenaar T. A., Melo M. N., Thallmair S., Carpenter T. S., Monticelli L., Tieleman D. P., Schiøtt B., Souza P. C. T., Marrink S. J.. The Martini 3 Lipidome: Expanded and Refined Parameters Improve Lipid Phase Behavior. ACS Cent. Sci. 2025;11:1598. doi: 10.1021/acscentsci.5c00755. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Bugnon M., Goullieux M., Röhrig U. F., Perez M. A. S., Daina A., Michielin O., Zoete V.. SwissParam 2023: A Modern Web-Based Tool for Efficient Small Molecule Parametrization. J. Chem. Inf. Model. 2023;63(21):6469–6475. doi: 10.1021/acs.jcim.3c01053. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Zoete V., Cuendet M. A., Grosdidier A., Michielin O.. SwissParam: A Fast Force Field Generation Tool for Small Organic Molecules. J. Comput. Chem. 2011;32(11):2359–2368. doi: 10.1002/jcc.21816. [DOI] [PubMed] [Google Scholar]
  24. Alessandri R., Souza P. C. T., Thallmair S., Melo M. N., de Vries A. H., Marrink S. J.. Pitfalls of the Martini Model. J. Chem. Theory Comput. 2019;15(10):5448–5460. doi: 10.1021/acs.jctc.9b00473. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Daina A., Michielin O., Zoete V.. SwissADME: A Free Web Tool to Evaluate Pharmacokinetics, Drug-Likeness and Medicinal Chemistry Friendliness of Small Molecules. Sci. Rep. 2017;7:42717. doi: 10.1038/srep42717. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Cheng T., Zhao Y., Li X., Lin F., Xu Y., Zhang X., Li Y., Wang R., Lai L.. Computation of Octanol–Water Partition Coefficients by Guiding an Additive Model with Knowledge. J. Chem. Inf. Model. 2007;47(6):2140–2148. doi: 10.1021/ci700257y. [DOI] [PubMed] [Google Scholar]
  27. Vainikka P., Thallmair S., Souza P. C. T., Marrink S. J.. Martini 3 Coarse-Grained Model for Type III Deep Eutectic Solvents: Thermodynamic, Structural, and Extraction Properties. ACS Sustainable Chem. Eng. 2021;9(51):17338–17350. doi: 10.1021/acssuschemeng.1c06521. [DOI] [Google Scholar]
  28. Müller C., Zhang L., Zipfel S., Topitsch A., Lutz M., Eckert J., Prasser B., Chami M., Lü W., Du J., Einsle O.. Molecular Interplay of an Assembly Machinery for Nitrous Oxide Reductase. Nature. 2022;608(7923):626–631. doi: 10.1038/s41586-022-05015-2. [DOI] [PubMed] [Google Scholar]
  29. Presutti, C. ; Bozoflu, M. ; Juhaniewicz-Dębińska, J. ; Lowe, L. A. ; Woltinge, N. ; Das, A. ; Stuart, M. ; Marrink, S. J. ; Poolman, B. . The Physicochemical Design, Structure and Permeability of the JCVI-syn3A Minimal Cell Membrane, 2026, 10.21203/rs.3.rs-8843141/v1. [DOI]
  30. Hilpert C., Beranger L., Souza P. C. T., Vainikka P. A., Nieto V., Marrink S. J., Monticelli L., Launay G.. Facilitating CG Simulations with MAD: The MArtini Database Server. J. Chem. Inf. Model. 2023;63(3):702–710. doi: 10.1021/acs.jcim.2c01375. [DOI] [PubMed] [Google Scholar]
  31. Barriga, R. Simulating Substrate Binding Sites in the S. aureus Type II NADH Dehydrogenase Universidade Nova de Lisboa, Instituto de Tecnologia Química e Biológica António Xavie; 2023. [Google Scholar]
  32. Szczuka, M. ; Pereira, G. P. ; Walter, L. J. ; Gueroult, M. ; Poulain, P. ; Bereau, T. ; De Souza, P. T. ; Chavent, M. . Fast Parameterization of Martini3Models for Fragments and Small Molecules, J. Chem. Theory Comput. 2026. 22 610–623 10.1101/2025.07.13.6 [DOI] [PubMed] [Google Scholar]
  33. Kelidou M., Stroh K. S., Risselada H. J.. Automated Parametrization of Small Molecules within the Martini 3 Coarse-Grained Model Guided by Experimental Log P Values. Sci. Rep. 2025;15(1):37169. doi: 10.1038/s41598-025-24757-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Bonneau K., Pasos-Trejo A. S., Plainer M., Sagresti L., Venturin J., Zaporozhets I., Caruso A., Rolando E., Guljas A., Klein L., Schebek M., Albani F., López-Ríos de Castro R., El Machachi Z., Giambagli L., Clementi C.. Breaking the Barriers of Molecular Dynamics With Deep-Learning: Opportunities, Pitfalls, and How to Navigate Them. WIREs Comput. Mol. Sci. 2026;16(1):e70064. doi: 10.1002/wcms.70064. [DOI] [Google Scholar]
  35. Yesselman J. D., Price D. J., Knight J. L., Brooks C. L.. MATCH: An Atom-Typing Toolset for Molecular Mechanics Force Fields. J. Comput. Chem. 2012;33(2):189–202. doi: 10.1002/jcc.21963. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Parrinello M., Rahman A.. Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method. J. Appl. Phys. 1981;52(12):7182–7190. doi: 10.1063/1.328693. [DOI] [Google Scholar]
  37. Bernetti M., Bussi G.. Pressure Control Using Stochastic Cell Rescaling. J. Chem. Phys. 2020;153(11):114107. doi: 10.1063/5.0020514. [DOI] [PubMed] [Google Scholar]
  38. Bussi G., Donadio D., Parrinello M.. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007;126(1):014101. doi: 10.1063/1.2408420. [DOI] [PubMed] [Google Scholar]
  39. De Jong D. H., Baoukina S., Ingólfsson H. I., Marrink S. J.. Martini Straight: Boosting Performance Using a Shorter Cutoff and GPUs. Comput. Phys. Commun. 2016;199:1–7. doi: 10.1016/j.cpc.2015.09.014. [DOI] [Google Scholar]
  40. Kim H., Fábián B., Hummer G.. Neighbor List Artifacts in Molecular Dynamics Simulations. J. Chem. Theory Comput. 2023;19(23):8919–8929. doi: 10.1021/acs.jctc.3c00777. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Grünewald, F. ; Marrink, S. J. . Fgrunewald/Fast_forward, 2022. 10.5281/zenodo.6531922 [DOI]
  42. Barnoud, J. CG Builder. https://jbarnoud.github.io/cgbuilder/.
  43. Grünewald F., Seute L., Alessandri R., König M., Kroon P. C.. CGsmiles: A Versatile Line Notation for Molecular Representations across Multiple Resolutions. J. Chem. Inf. Model. 2025;65(7):3405–3419. doi: 10.1021/acs.jcim.5c00064. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Brasnett, C. ; Fidlin, M. ; Duve, T. ; Thallmair, S. ; Marrink, S. J. ; Grünewald, F. . Fast-Forward: Automatic Assignment and Assessment of Bonded Parameters for the Martini Force Field, 2026. , 10.26434/chemrxiv.15001072/v1 [DOI]
  45. Grünewald F., Alessandri R., Kroon P. C., Monticelli L., Souza P. C. T., Marrink S. J.. et al. Polyply; a Python Suite for Facilitating Simulations of Macromolecules and Nanomaterials. Nat. Commun. 2022;13(1):68. doi: 10.1038/s41467-021-27627-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Goga N., Rzepiela A. J., De Vries A. H., Marrink S. J., Berendsen H. J. C.. Efficient Algorithms for Langevin and DPD Dynamics. J. Chem. Theory Comput. 2012;8(10):3637–3649. doi: 10.1021/ct3000876. [DOI] [PubMed] [Google Scholar]
  47. Shirts M. R., Chodera J. D.. Statistically Optimal Analysis of Samples from Multiple Equilibrium States. J. Chem. Phys. 2008;129(12):124105. doi: 10.1063/1.2978177. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Wu Z., Dotson D. L., Alibay I., Allen B. K., Barhaghi M. S., Hénin J., Joseph T. T., Kenney I. M., Lee H., Li H., Lim V., Liu S., Marson D., Merz P. T., Schlaich A., Mobley D., Shirts M. R., Beckstein O.. Alchemlyb: The Simple Alchemistry Library. J. Open Source Software. 2024;9(101):6934. doi: 10.21105/joss.06934. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Kroon P. C., Grunewald F., Barnoud J., van Tilburg M., Brasnett C., de Souza P. C. T., Wassenaar T. A., Marrink S.-J. J.. Martinize2 and Vermouth: Unified Framework for Topology Generation. eLife. 2025;12:e90627. doi: 10.7554/eLife.90627.3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Wassenaar T. A., Ingólfsson H. I., Böckmann R. A., Tieleman D. P., Marrink S. J.. Computational Lipidomics with Insane: A Versatile Tool for Generating Custom Membranes for Molecular Simulations. J. Chem. Theory Comput. 2015;11(5):2144–2155. doi: 10.1021/acs.jctc.5b00209. [DOI] [PubMed] [Google Scholar]
  51. Song W., Corey R. A., Ansell T. B., Cassidy C. K., Horrell M. R., Duncan A. L., Stansfeld P. J., Sansom M. S. P.. PyLipID: A Python Package for Analysis of Protein–Lipid Interactions from Molecular Dynamics Simulations. J. Chem. Theory Comput. 2022;18(2):1188–1201. doi: 10.1021/acs.jctc.1c00708. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Humphrey W., Dalke A., Schulten K.. VMD: Visual Molecular Dynamics. J. Mol. Graphics. 1996;14(1):33–38. doi: 10.1016/0263-7855(96)00018-5. [DOI] [PubMed] [Google Scholar]
  53. Hunter J. D.. Matplotlib: A 2D Graphics Environment. Comput. Sci. Eng. 2007;9(3):90–95. doi: 10.1109/MCSE.2007.55. [DOI] [Google Scholar]
  54. Brasnett C., Marrink S. J.. MartiniGlass: A Tool for Enabling Visualization of Coarse-Grained Martini Topologies. J. Chem. Inf. Model. 2025;65(7):3137–3141. doi: 10.1021/acs.jcim.4c02277. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Hummer G., Szabo A., Chandler D.. Free Energy Reconstruction from Nonequilibrium Single-Molecule Pulling Experiments. Proc. Natl. Acad. Sci. U.S.A. 2001;98(7):3658–3661. doi: 10.1073/pnas.071034098. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Lee C. T., Comer J., Herndon C., Leung N., Pavlova A., Swift R. V., Tung C., Rowley C. N., Amaro R. E., Chipot C., Wang Y., Gumbart J. C.. Simulation-Based Approaches for Determining Membrane Permeability of Small Compounds. J. Chem. Inf. Model. 2016;56(4):721–733. doi: 10.1021/acs.jcim.6b00022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Marrink S.-J., Berendsen H. J. C.. Simulation of Water Transport through a Lipid Membrane. J. Phys. Chem. A. 1994;98(15):4155–4168. doi: 10.1021/j100066a040. [DOI] [Google Scholar]
  58. Brasnett, C. ; Brown, C. ; Stevens, J. A. ; Grünewald, L. ; Marrink, S. . Supporting Information for the Martini 3 Metabolome, 2026. 10.5281/zenodo.18622430 [DOI] [PMC free article] [PubMed]
  59. Marin, R. ; Hilpert, C. ; Grünewald, F. ; Valério, M. ; Borges-Araúj, L. ; Janczarski, S. ; Rossini, N. O. ; Marrink, S. J. ; Souza, P. C. T. ; Launay, G. . Extending the MAD Toolbox: New Polymer Builder and Enhanced Martini Database, 2026, 10.64898/2026.01.23.700524 . [DOI]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

ct6c00463_si_001.csv (104.2KB, csv)
ct6c00463_si_002.pdf (887.9KB, pdf)

Data Availability Statement

Full Supporting Information is available on Zenodo. A subset of Supporting Information (CG topology and coordinate files, mapping files, and bonded distribution comparisons) is available on the M3-Metabolome repository on GitHub: https://github.com/Martini-Force-Field-Initiative/M3-Metabolome. Parameters will also be made available on the MArtini Database (MAD) web server. ,

Articles from Journal of Chemical Theory and Computation are provided here courtesy of American Chemical Society

RESOURCES