Published December 2, 2025 | Version v6

Repository of Auto-MartiniM3: Fast parametrization of Martini3 models for fragments and small molecules

  • 1. ROR icon Université de Toulouse
  • 2. CALIXAR Eurofins
  • 3. CNRS Délégation Alpes
  • 4. ROR icon Heidelberg University
  • 5. ROR icon Laboratoire de Microbiologie et Génétique Moléculaires
  • 6. ROR icon Centre National de la Recherche Scientifique
  • 7. Université Paris Cité
  • 8. École Normale Supérieure de Lyon

Description

Here, we introduce the data repository of Auto-MartiniM3, an advanced and updated version of the Auto-Martini program, designed to automate the coarse-graining of small molecules to be used with the Martini 3 force field.

We validated our approach by modeling 81 small molecules from the Martini Database and comparing their water-oil partitionning behavior during solvation free energy simulations with ones obtained from models designed by Martini experts. - simulation and TI computation results, as well as coarse-grain parametrizations created with Auto-MartiniM3 are avaliable in TI_simulations.zip

Additionally, we assessed the behavior of Auto-MartiniM3-generated models by calculating solute translocation and free energy across lipid bilayers. Computation and simulation data and Auto-MartiniM3-made models (with some manually introduced changes for increasing stability) are available in PMF_simulations.zip

We evaluated Auto-MartniM3-made caffeine model by testing its binding to the adenosine A2A receptor. Simulation data are available in A2A_POPC_AM3caffeine.zip

To confirm applicability of Auto-MartiniM3 as the tool for high-throughput coarse grain parametrization, we deployed it on 100 000 molecules from Enamine Fragment Collection (files avaliable in Enamine100k_parametrization.zip) and simulated 23619 ligands in a box of water (simulation results are avaliable in Enamine24k_simulations.zip).

Auto-MartiniM3 source code is freely available at https://github.com/Martini-Force-Field-Initiative/Automartini_M3

Methods

Solvation free energy simulations : Systems are energy minimized using the steepest descent algorithm for a maximum of 10,000 steps with LINCS being used to control constraints. A short equilibration step follows, for 5 ns with an integration timestep of 10 fs and with the Verlet algorithm as a cut-off scheme (straight cut-off at 1.1 nm). The temperature is controlled using the velocity-rescale algorithm and set at 298K and pressure is controlled using the Berendsen barostat with a compressibility of 3 ⋅ 10^-4 bar^-1 and a coupling parameter of 4.0 ps at 1 bar pressure. The solvation free energies are computed using thermodynamic integration, an alchemical free energy method. For each system, a series of 11 simulations with equally spaced λ windows are carried out with a stochastic dynamics integrator with a 20 fs timestep and employing the Parrinello-Rahman barostat and a compressibility at 1 bar with coupling parameter of 4.0 ps. Each window runs for 5 ns, using a soft-core potential while setting α = 0.5 and power set to 1. The Bennet Acceptable Ratio is used to compute the free energies and the associated error. The free energy of transferring a given molecule from water to one of the other solvents (OCT, HD, CLF) is computed as the difference between the solvation free energy of that molecule in each solvent (ΔΔGOCT-W = ΔGvac-OCT - ΔGvac-W). Simulations are carried out using GROMACS 2023.2.

Potential of Mean Force of molecule insertion in lipid membrane : We use umbrella sampling to calculate the PMF of the molecule insertion. The collective variable employed is the distance to the bilayer center, projected on the bilayer normal. This yields 92 umbrella windows, with a distance to the bilayer center between 0 and 3.95 nm. Each umbrella system is energy minimized using a steepest descent algorithm. This step is followed by two equilibration simulations of 100 ps and 2 ns with integration timesteps 10 fs and 20 fs, respectively. Finally, each umbrella window is subjected to a 120-ns production simulation with a 20-fs timestep and a harmonic biasing potential of 1000 kJ/mol/nm2. The temperature of the system is maintained at 297K using the velocity-rescale thermostat. The pressure is set to 1 bar and controlled using a semi-isotropic Parrinello-Rahman barostat with a compressibility of 3 ⋅ 10^-4 bar^-1. Regarding the non-bonded interactions, we employ a cutoff of 1.1 nm in conjunction with a potential shift. For electrostatic interactions, a reaction field with a dielectric constant of 15 is used. To prevent membrane artifacts we manually set the rlist parameter to 1.35 nm. The simulations are carried out using GROMACS 2024.2. To obtain the final PMF profiles, we use the weighted histogram analysis method (WHAM).

Interaction of caffeine with A2A receptor embedded in POPC membrane : The molecular dynamics simulations are performed in GROMACS 2023.2. For minimization, we use the steepest descent algorithm with the LINCS constraints algorithm. System is equilibrated in four stages, with integration steps of 2 fs, 5 fs, 10 fs and finishing with an integration step of 20 fs, for 10 ns at each stage and with the Verlet algorithm as a cut-off scheme with the straight cut-off at 1.1 nm. The temperature is controlled using the velocity-rescale algorithm and set at 320 K. The pressure is controlled using the Parrinello-Rahman barostat with a compressibility of 3 ⋅ 10^-4 bar^-1 and a coupling parameter of 12.0 ps, and a semi-isotropic pressure at 1 bar. We run 12 production runs of 20 microseconds, applying a 20-fs integration step using the same thermostat and barostat algorithms as done in the equilibration step.

High-throughput parametrization and simulation : Each ligand was embedded in a water box and energy-minimized using the steepest descent algorithm. The equilibration phase was carried out in three steps, during which the integration time step was progressively increased (2 fs, 10 fs, and 20 fs), and positional restraints were gradually removed. Production molecular dynamics simulations were then conducted for 100 ns with a 20 fs integration time step. The temperature was maintained at 310 K using a velocity-rescale thermostat, while pressure was regulated at 1 bar using a c-rescale barostat (84), with a compressibility of 3⋅10−4 bar-1 and a coupling constant of 4.0 ps.

Files

A2A_POPC_AM3caffeine.zip

Files (37.2 GB)

Name Size
md5:74e424c07b3e5fef2559a5a4e27a0701
8.2 GB Preview Download
md5:d5942a3587771c4dff6b3a2bb7c98b92
144.2 MB Preview Download
md5:0a2aa41fec200078ebd99287ea6d8a76
3.6 MB Preview Download
md5:c90b26ea51054f628a1f31746dc78d0e
14.1 GB Preview Download
md5:a49e904e5e58970a2dcca321887d5ef1
14.7 GB Preview Download

Additional details

Dates

Updated
2025-10-30