Overview
Membrane proteins are the richest class of drug targets in pharmacology: GPCRs alone are the targets of about 36% of approved drugs,[1] and ion channels, transporters, and receptors of every kind carry their biology at a lipid interface. Simulating any of them, whether to validate a ligand, study a mutation, or watch a binding event, requires something no experiment provides: a complete atomistic assembly of the protein embedded in a lipid bilayer, solvated and ionized, with correct covalent chemistry and no packing artifacts. Crystallography removes the membrane; the modeler has to put it back.
Putting it back is genuinely difficult. The protein must sit in the membrane at the right tilt and depth, which is itself a measurable quantity. The lipids must pack around it without threading through aromatic rings or sterol rings, defects that no amount of molecular dynamics can repair. Every titratable residue needs a protonation state, and the buried, charged ones can depend on the ligands present. The bilayer must start at its equilibrium density or drift for hundreds of nanoseconds. And real membranes are asymmetric: the outer leaflet of a plasma membrane is rich in sphingomyelin and cholesterol while the inner leaflet carries the phosphatidylserine and phosphatidylethanolamine, and that asymmetry is biology, not decoration.
Autochem's membrane construction protocol builds the whole system in house, from a protein structure file to a quality-gated, simulation-ready bundle. The protein is protonated and oriented; a coarse-grained Martini 3 bilayer with the requested per-leaflet composition self-packs around it; the equilibrated assembly is backmapped to all-atom resolution with a conformer library built from our own simulations; the system is parameterized (ff19SB protein, Lipid21 lipids, OPC water), solvated, ionized, minimized, and passed through a battery of structural quality gates before anything is called a build. Every build ships with a QC report recording what each stage did and measured.
The protocol runs entirely on our own infrastructure, against published science: the assembly strategy follows the CHARMM-GUI Membrane Builder lineage[17][19] and the coarse-grained assembly idea behind MemProtMD,[9] and the builder is validated against scattering and NMR experiments, never against another simulation.
Scientific Background
Why geometric placement is not enough
The classical way to build a membrane system is geometric: place lipid conformers on a grid around the protein, delete the clashes, minimize. This is fast, but the annular shell, the first layer of lipids touching the protein, ends up an artifact of the packing algorithm rather than a physical structure. GPCRs carry defined cholesterol grooves; channels and transporters thin the membrane locally; an annular shell has to be discovered by the lipids, not assumed. Placement also concentrates the failure modes that survive into production: a lipid tail threaded through a tryptophan or sterol ring is covalently topologically wrong, and no minimization or MD run undoes it.[18][19]
Assembly by coarse-grained simulation, then backmapping
The protocol's core idea is to let physics place the lipids. The protein is converted to a Martini 3 coarse-grained representation with martinize2,[5][6] and a bilayer with the exact requested composition is generated around it with insane[8] and equilibrated under NPT with the protein restrained. Over tens of nanoseconds of coarse-grained dynamics the lipids find their own contacts with the protein: annular shells, cholesterol binding grooves, local thinning. The equilibrated coarse-grained frame is then rebuilt at all-atom resolution. This multiscale strategy is the state of the art for lipid-protein contacts[9] and the backmapping step follows the fragment- and conformer-based lineage of backward and CG2AT2,[10] implemented here against the Lipid21[11] residue vocabulary with our own conformer library.
Asymmetry must be engineered
A symmetric bilayer would be simpler, but it would not be a plasma membrane. Real membranes keep different lipid compositions in their two leaflets, maintained by active cellular machinery, and that asymmetry matters to drug discovery science: the polybasic inner leaflet recruits signaling proteins, sphingomyelin and cholesterol order the outer leaflet, and the resulting differential stress bends and regulates membranes. A building protocol that scrambles leaflets during assembly, which is exactly what self-assembly from a random soup does, destroys the requested asymmetry before the first all-atom frame. The protocol treats leaflet identity as a first-class constraint: the coarse-grained bilayer is generated pre-formed with the two requested compositions, and every lipid is restrained to, and measured in, its own leaflet throughout (see Asymmetric Membranes).
Pipeline Overview
The membrane construction workflow consists of eight stages:
- Protein preparation: pH-correct protonation, disulfides, covalent cofactors and ligands
- Orientation: placing the protein in the membrane frame (OPM or transfer-energy fit)
- Coarse-grained assembly: Martini 3 bilayer built and equilibrated around the protein
- Backmapping: all-atom reconstruction from the conformer library
- Solvation and ions: OPC water and SLTCAP ion counting
- Parameterization and export: one consistent AMBER-family topology
- Minimization and quality gates: soft-core minimization and ten structural checks
- Handoff: a simulation-ready bundle for equilibration and production MD
Stage 1: Protein Preparation
The input structure is prepared for the AMBER force field family. PDB2PQR with PROPKA[15] assigns residue protonation states at the system pH and writes AMBER residue names; hydrogens are rebuilt from templates, disulfides are detected from geometry and bonded, and chain breaks are reported rather than silently joined.
Two pieces of chemistry go beyond the standard preparation, because membrane proteins carry them constantly:
- Covalent cofactors are kept, bonded, and parametrized. S-palmitoylated cysteines (a thioester on a membrane-proximal cysteine, common in GPCRs) are rebuilt from Lipid21's own palmitoyl tail parameters and a thioester carbonyl unit, so the chain behaves like the membrane's own lipid. Retinal on its lysine, the protonated Schiff base of every rhodopsin, is parametrized with a custom iminium type that keeps the C=N double bond a double bond, preserving the crystallographic isomer (11-cis in dark rhodopsin, all-trans in microbial rhodopsins).
- Ligands and ions are kept with explicit chemistry. Organic ligands and cofactors get GAFF2 parameters with AM1-BCC charges; their crystal heavy atoms keep their coordinates. Crystal ions are recognized and given their OPC-paired ion parameters. Waters, detergents, and crystallization additives are dropped and every decision is recorded in the QC report.
When ligands are present, protonation sees them: PROPKA runs twice on the same protein, without and with the kept het groups, and any residue whose state differs takes the with-ligand state. This isolates the ligands' effect on, for example, a Walker-B glutamate in an ATP-binding site, which no protein-only calculation can see.
Stage 2: Orientation in the Membrane Frame
Everything downstream assumes the membrane normal is z and the bilayer midplane is z = 0, so the protein must be placed there first, at the correct tilt and depth. Two routes, in order of preference:
- OPM lookup.[2] The Orientations of Proteins in Membranes database holds literature-standard placements for membrane proteins of known structure. The prepared structure is superposed onto the OPM entry's oriented coordinates by sequence alignment. For models and renamed uploads whose own PDB id OPM does not know, the builder finds the protein's structures by RCSB sequence search and superposes the OPM entries of those.
- Transfer-energy slab fit, for anything OPM lacks (de novo models, designs). Following the PPM approach,[3][4] the protein is scored in a hydrophobic slab with a smooth water-to-hydrocarbon interface: the transfer free energy of each residue's exposed surface, using the Wimley-White whole-residue water-to-octanol scale, is minimized over the membrane normal, the shift along it, and the slab half-thickness.
Both routes are validated against OPM itself: ten membrane proteins (GPCRs, channels, transporters, and beta-barrels), each randomly displaced and re-fitted blind, land within a mean 1.9° of OPM's membrane normal with midplanes within 1.4 Å; seven ABC transporters, whose large soluble domains pull the centroid far from the membrane and are the hardest case for any fit, land within 10° and 2 Å of their OPM placements.
Stage 3: Coarse-Grained Assembly
The oriented protein is coarse-grained with martinize2[5] (Martini 3,[6] elastic network, disulfides detected), and insane[8] generates a pre-formed bilayer around it with the exact requested composition per leaflet, using the 2025 Martini 3 lipidome parameters.[7] Starting from a pre-formed bilayer rather than random self-assembly is what makes an asymmetric build possible: self-assembly scrambles the leaflets before the simulation can preserve them.
The system then runs tens of nanoseconds of NPT coarse-grained dynamics (50 ns by default) with every protein bead position-restrained. This is the stage where the membrane earns its structure: lipids pack against the protein's actual surface, cholesterol finds its grooves, the bilayer thins where the protein is narrow, and the whole assembly settles to its equilibrium area per lipid. The coarse-grained representation makes this cheap; the throughput of these small systems is in the thousands of nanoseconds per day on ordinary CPU hardware.
Leaflet identity is enforced and measured throughout: each lipid stays in the leaflet it was generated in, free to move and undulate within it but not to flip across, and each lipid's leaflet is measured before and after the coarse-grained run. A lipid ending in a leaflet that requested none of it is a QC finding, and the final topology lists lipids by their measured leaflet. The protein is also re-centered on its transmembrane core, so soluble domains on one side do not push the membrane off the box center.
Stage 4: Backmapping to All-Atom
The equilibrated coarse-grained frame is rebuilt at all-atom resolution in two moves. The protein is the prepared all-atom input itself, rigidly fitted onto its own coarse-grained backbone, so no atomistic detail is ever invented for it. Each lipid takes, from a conformer library, the all-atom conformer whose pseudoatom centroids best fit its coarse-grained beads, placed rigidly by Kabsch superposition with proper rotations only, so stereocentres cannot invert and every bond, angle, and cis double bond arrives from an equilibrated simulation rather than a template.
The library is generated by us, from our own equilibrium simulations of every supported lipid. It currently holds 83,336 conformers covering the 82 supported lipids (80 glycerophospholipids, palmitoyl sphingomyelin, and cholesterol), so every placed molecule arrives with physically realistic internal geometry.
Backmapping is where ring threading is caught and fixed. A lipid bond threaded through an aromatic or sterol ring cannot be repaired later, so every placement is checked, and an offending lipid is re-placed with its next-best conformer until none threads. In the reference builds this catches and fixes the handful of lipids that would otherwise silently corrupt a production run.
Stage 5: Solvation and Ions
The system is solvated with OPC[13] water, a 4-point model chosen for its accuracy in reproducing hydration thermodynamics. Water is excluded from the hydrocarbon core, with one physically important exception: where the coarse-grained equilibration had water (a channel pore, an open transporter cavity, a hydrated funnel), all-atom water is kept there, because deleting it would leave a vacuum the lipids would collapse into.
Ion counts come from SLTCAP,[16] which solves for the cation and anion numbers that produce the requested bulk concentration around a charged solute. The naive rule (neutralize, then add concentration times volume) overshoots the bulk concentration because the counterions themselves screen; SLTCAP is the corrected count and always returns an exactly neutral system. Ions take the Li/Merz parameters fitted to OPC,[14] and their placement is seeded, so a build is reproducible.
Stage 6: Parameterization and Topology Export
The whole system, protein, covalent cofactors, lipids, ligands, ions, water, is parametrized in one consistent AMBER-family stack: ff19SB protein,[12] Lipid21 lipids,[11] OPC water, and GAFF2 for ligands and custom units, assembled by tleap into a single prmtop with disulfides and cofactor links bonded explicitly.
Export to GROMACS is verified rather than assumed: the standard converter mishandles this particular force-field combination in ways that silently distort its energetics, so the export is our own, and it is checked the only way that counts, a term-by-term single-point energy comparison against the AMBER reference on the same coordinates:
| Energy term (kJ/mol) | This pipeline | AMBER reference |
|---|---|---|
| Angles | 98253.47 | 98253.47 |
| Proper + improper dihedrals | 73240.09 | 73240.14 |
| CMAP corrections | 1488.583 | 1488.584 |
| 1-4 van der Waals | 36229.88 | 36229.88 |
| 1-4 Coulomb | -46482.88 | -46482.86 |
Equilibration restraints are exported with the bundle in the CHARMM-GUI convention[20]: position restraints on protein backbone and side chains, membrane-headgroup z restraints on lipids, and dihedral restraints holding glycerol chirality and every cis double bond, with target angles verified against thousands of equilibrated conformers.
Stage 7: Minimization and Quality Gates
The assembled system is minimized in two passes, first with soft-core van der Waals decoupling so that the dense, just-backmapped bilayer has finite forces from the first step, then with a plain minimization to convergence. The bundle is then measured, and only the unambiguous failures block a build:
| Quality gate | What it checks | On failure |
|---|---|---|
| Charged residues in the core | Dry charged groups (Lys, Arg, Asp, Glu) buried in the hydrocarbon core, the signature of a misoriented protein | Build fails |
| Ring penetration | No lipid bond threaded through an aromatic or sterol ring | Build fails |
| Clashes | No heavy atoms of different molecules overlapping after minimization | Build fails |
| Leaflet composition | No lipid in a leaflet that requested none of it; fractions vs the request | Build fails |
| Net charge | The system is exactly neutral | Build fails |
| Core vacuum | No vacuum bubble in the hydrocarbon core | Build fails |
| Orientation | Membrane tilt and midplane in the final system vs the chosen frame | Warning |
| Area per lipid | Per-leaflet area per molecule in a physical range | Warning |
| Salt concentration | Bulk ion concentration vs the request | Warning |
| Core water | No stray water in the core away from the protein | Warning |
Any failure that depends on the random placement seed triggers a full rebuild with a new seed, up to three attempts, each attempt uploading its partial report. Everything the build measured, stage timings, leaflet counts, dropped het groups, gate results, ends up in a QC report shipped with the bundle, along with a manifest of the exact tool and parameter-file versions used.
Stage 8: Handoff to Simulation
The build ships as a bundle in the CHARMM-GUI layout (coordinates, per-molecule topologies, index groups including the two leaflets, and the restraint definitions), ready for the equilibration, production, and analysis pipelines without modification. Equilibration ramps the position and dihedral restraints down stage by stage; production then runs the science. The built system is also rendered in the platform's structure viewer, membrane and both leaflets selectable, before any simulation is spent on it.
Asymmetric Membranes and Organelle Presets
The protocol's defining capability is the asymmetric build: each leaflet gets its own composition, specified in mole percent, and the two leaflets are balanced by area rather than lipid count, so different counts per leaflet are expected and correct. A plasma-membrane build, for example, may carry sphingomyelin and heavy cholesterol in the outer leaflet and phosphatidylserine in the inner one; the asymmetry survives assembly because the coarse-grained bilayer is generated pre-formed per leaflet, every lipid is z-restrained to its own leaflet during equilibration, and leaflet membership is measured again on the final system. Per-leaflet index groups let every downstream analysis separate the two leaflets' behavior, which is the point of building them separately.
Composition does not have to be invented from scratch. The agent reads the protein's annotated subcellular location from UniProt and pre-fills a biologically reasoned composition for it:
| Preset | Character | Example composition |
|---|---|---|
| Plasma membrane | Cholesterol-rich; SM outer, PS and PE inner | POPC 35, POPE 20, POPS 5, PSM 10, CHL 30 |
| Neuronal plasma membrane | Higher cholesterol and PS than generic plasma | POPC 25, POPE 15, POPS 10, PSM 10, CHL 40 |
| Endoplasmic reticulum | PC/PE-rich, low cholesterol, no sphingomyelin | POPC 60, POPE 30, POPS 5, CHL 5 |
| Golgi apparatus | Intermediate between ER and plasma membrane | POPC 55, POPE 25, POPS 5, PSM 5, CHL 10 |
| Mitochondrial inner | PE-rich and anionic | POPC 40, POPE 35, POPG 20, CHL 5 |
| Lysosome / late endosome | Less cholesterol than plasma membrane | POPC 45, POPE 25, POPS 5, PSM 5, CHL 20 |
Every value is editable per leaflet in the composition picker before the build runs, and any supported lipid mixture can be specified directly. The builder covers the 80 glycerophospholipids of the Lipid21 + Martini 3 lipid space (PC, PE, PS, PA, PG headgroups with saturated and unsaturated chains from 12 to 22 carbons), plus palmitoyl sphingomyelin and cholesterol.
Validation and Benchmarks
The builder is validated internally against experiment, never against another simulation. Three layers of evidence:
Built systems start at equilibrium
Because the coarse-grained stage equilibrates the bilayer at its own density before backmapping, the all-atom system starts equilibrated instead of relaxing toward equilibrium for hundreds of nanoseconds. On the reference system (a GPCR in a plasma-like mixture), the area per lipid after the builder's equilibration sits within about 2% of the same composition's value from a 450-nanosecond all-atom run, and the box area is flat to within 0.1% per nanoseconds in validation runs, with zero constraint-solver warnings throughout.
Bilayer properties against scattering and NMR
Membrane-only builds of pure and mixed compositions are compared with experiment on the two quantities scattering measures directly, area per lipid and head-to-head thickness, and on the deuterium order parameter SCD = (3cos²θ − 1)/2, the NMR observable most sensitive to tail packing. First results at 20 nanoseconds per system:
| System | Area per lipid (Ų) | Head-head thickness (Å) | SCD plateau (sn-1 / sn-2) |
|---|---|---|---|
| POPC, 303 K | 65.1 (experiment 64.3 ± 1.3[21]) | 38.7 | -0.19 / -0.20 (experiment ≈ -0.20) |
| POPE, 310 K | 59.0 | 41.2 | -0.23 / -0.24 |
| POPC:CHL 70:30, 310 K | 50.6 per molecule | 43.0 (+4.3 vs pure POPC) | -0.27 / -0.29 |
| Asymmetric plasma-like, 310 K | 48.7 per molecule | 43.8 | -0.29 / -0.30 |
| Mitochondrial-like mix, 310 K | 58.6 | 41.0 | -0.23 / -0.24 |
The physics reads correctly across the set: cholesterol condenses the bilayer (smaller area per molecule, thicker membrane, higher tail order), the asymmetric plasma-like mix behaves like its outer-leaflet sterol content dictates, and both POPC tails show the expected dip at the cis double bond.
The benchmark: published experiments only
The standing benchmark is a set of systems where each input is a published experiment and each reference is what that experiment measured, with the pass tolerance stated next to every comparison:
| System | Compared against | Pass criterion |
|---|---|---|
| Pure POPC, DOPC, DMPC, POPE bilayers | Area per lipid and head-head thickness from X-ray / neutron scattering[21] | Within 3% of the published range |
| GWALP23 peptide in DLPC, DMPC, DOPC | 2H quadrupolar splittings of six labeled alanines (solid-state NMR)[22]: how the helix tilts as the membrane thins | RMSD ≤ 3 kHz over the six alanines |
| Influenza M2 channel in DMPC | Helix tilt from 15N solid-state NMR[23] | Within experimental error + 3° |
| β2-adrenergic receptor in POPC:CHL | The two cholesterol sites of its crystal structure[24]: crystal cholesterols removed, the membrane's own must find them | Each site occupied at ≥ 2× the no-binding baseline |
Every run in the benchmark also has to finish cleanly, zero constraint-solver warnings and box-area drift under 1% per nanosecond, before any physics comparison is scored. This is the standard every build is held to internally before the tool is trusted with a new system class.
Summary of Tools
| Stage | Tool | Purpose | Reference |
|---|---|---|---|
| Protonation | PDB2PQR + PROPKA | pH-correct states, ligand-aware | [15] |
| Protein force field | ff19SB (AmberTools tleap) | Protein parameters | [12] |
| Ligand parameters | GAFF2 + AM1-BCC | Cofactors and ligands | |
| Orientation | OPM database + PPM-style transfer-energy fit | Protein in the membrane frame | [2], [3], [4] |
| Coarse-graining | martinize2 (Martini 3) | Protein to pseudoatoms | [5], [6] |
| Bilayer generation | insane + Martini 3 lipidome (2025) | Pre-formed per-leaflet bilayer | [7], [8] |
| Coarse-grained MD | GROMACS | Lipid self-packing around the protein | |
| Backmapping | In-house conformer library (Lipid21 vocabulary) | All-atom reconstruction | [10], [11] |
| Water model | OPC | 4-point explicit solvent | [13] |
| Ions | SLTCAP counting + Li/Merz OPC parameters | Correct bulk concentration, neutral box | [14], [16] |
| Lipid force field | Lipid21 | Modular all-atom lipids | [11] |
| Topology export | AmberTools + ParmEd (with 1-4 and CMAP corrections) | One consistent GROMACS topology | |
| Minimization | GROMACS (soft-core + plain) | Overlap-free all-atom start | |
| Quality gates | In-house (ring penetration, leaflets, core checks) | Never ship a wrong system | [18], [19] |
| Bundle conventions | CHARMM-GUI layout | Direct handoff to equilibration and production | [17], [19], [20] |
| Composition presets | UniProt subcellular location | Organeller default mixtures |
Infrastructure
The builder runs as a container job on our own infrastructure: a typical GPCR-sized system builds end to end in about an hour, dominated by the coarse-grained equilibration. Builds are reproducible by seed (the same input and seed produce the same assembly), and every build's QC report carries a manifest of tool, library, and parameter-file versions. The output bundle follows the CHARMM-GUI layout, so the platform's equilibration, ligand insertion, production, and analysis pipelines run on it unchanged, and the built system is inspectable in the structure viewer before any compute is spent simulating it.
References
-
Hauser, A.S., Attwood, M.M., Rask-Andersen, M., Schiöth, H.B., Gloriam, D.E. (2018). Trends in GPCR drug discovery: new agents, targets and indications. Nature Reviews Drug Discovery, 16(12), 829-842. https://doi.org/10.1038/nrd.2017.178
-
Lomize, M.A., Pogozheva, I.D., Joo, H., Mosberg, H.I., Lomize, A.L. (2012). OPM database and PPM web server: resources for positioning of proteins in membranes. Nucleic Acids Research, 40(D1), D370-D376. https://doi.org/10.1093/nar/gkr703
-
Lomize, A.L., Pogozheva, I.D., Lomize, M.A., Mosberg, H.I. (2006). Positioning of proteins in membranes: a computational approach. Protein Science, 15(6), 1318-1333. https://doi.org/10.1110/ps.062126106
-
Lomize, A.L., Todd, S.C., Pogozheva, I.D. (2022). Spatial arrangement of proteins in planar and curved membranes by PPM 3.0. Protein Science, 31(1), 209-220. https://doi.org/10.1002/pro.4246
-
Kroon, P.C. et al. (2025). Martinize2 and Vermouth: unified framework for topology generation. eLife. https://doi.org/10.7554/eLife.90627
-
Souza, P.C.T. et al. (2021). Martini 3: a general purpose force field for coarse-grained molecular dynamics. Nature Methods, 18(4), 382-388. https://doi.org/10.1038/s41592-021-01098-3
-
Pedersen, K.B. et al. (2025). The Martini 3 lipidome: expanded and refined parameters improve lipid phase behavior. ACS Central Science. https://doi.org/10.1021/acscentsci.5c00755
-
Wassenaar, T.A., Ingólfsson, H.I., Böckmann, R.A., Tieleman, D.P., Marrink, S.J. (2015). Computational lipidomics with insane: a versatile tool for generating custom membranes for molecular simulations. Journal of Chemical Theory and Computation, 11(5), 2144-2155. https://doi.org/10.1021/acs.jctc.5b00209
-
Newport, T.D., Sansom, M.S.P., Stansfeld, P.J. (2019). The MemProtMD database: a resource for membrane-embedded protein structures and their lipid interactions. Nucleic Acids Research, 47(D1), D390-D397. https://doi.org/10.1093/nar/gky1047
-
Vickery, O.N., Stansfeld, P.J. (2021). CG2AT2: an enhanced fragment-based approach for serial multi-scale molecular dynamics simulations. Journal of Chemical Theory and Computation, 17(10), 6472-6482. https://doi.org/10.1021/acs.jctc.1c00295
-
Dickson, C.J., Walker, R.C., Gould, I.R. (2022). Lipid21: complex lipid membrane simulations with AMBER. Journal of Chemical Theory and Computation, 18(4), 1726-1736. https://doi.org/10.1021/acs.jctc.1c01217
-
Tian, C., Kasavajhala, K., Belfon, K.A.A., Raguette, L., Huang, H., Migues, A.N., Bickel, J., Wang, Y., Pincay, J., Wu, Q., Simmerling, C. (2020). ff19SB: amino-acid-specific protein backbone parameters trained against quantum mechanics energy surfaces in solution. Journal of Chemical Theory and Computation, 16(1), 528-552. https://doi.org/10.1021/acs.jctc.9b00591
-
Izadi, S., Anandakrishnan, R., Onufriev, A.V. (2014). Building water models: a different approach. Journal of Physical Chemistry Letters, 5(21), 3863-3871. https://doi.org/10.1021/jz501780a
-
Sengupta, A., Li, Z., Song, L.F., Li, P., Merz, K.M. (2021). Parameterization of monovalent ions for the OPC3, OPC, TIP3P-FB, and TIP4P-FB water models. Journal of Chemical Information and Modeling, 61(2), 869-880. https://doi.org/10.1021/acs.jcim.0c01390
-
Olsson, M.H.M., Søndergaard, C.R., Rostkowski, M., Jensen, J.H. (2011). PROPKA3: consistent treatment of internal and surface residues in empirical pKa predictions. Journal of Chemical Theory and Computation, 7(2), 525-537. https://doi.org/10.1021/ct100578z
-
Schmit, J.D., Kariyawasam, N.L., Needham, V., Smith, P.E. (2018). SLTCAP: a simple method for calculating the number of ions needed for MD simulation. Journal of Chemical Theory and Computation, 14(4), 1823-1827. https://doi.org/10.1021/acs.jctc.7b01254
-
Jo, S., Kim, T., Im, W. (2007). Automated builder and database of protein/membrane complexes for molecular dynamics simulations. PLoS ONE, 2(10), e880. https://doi.org/10.1371/journal.pone.0000880
-
Jo, S., Lim, J.B., Klauda, J.B., Im, W. (2009). CHARMM-GUI Membrane Builder for mixed bilayers and its application to yeast membranes. Biophysical Journal, 97(1), 50-58. https://doi.org/10.1016/j.bpj.2009.04.013
-
Wu, E.L. et al. (2014). CHARMM-GUI Membrane Builder toward realistic biological membrane simulations. Journal of Computational Chemistry, 35(27), 1997-2004. https://doi.org/10.1002/jcc.23702
-
Lee, J. et al. (2016). CHARMM-GUI Input Generator for NAMD, GROMACS, AMBER, OpenMM and CHARMM/OpenMM simulations using the CHARMM36 additive force field. Journal of Chemical Theory and Computation, 12(1), 405-413. https://doi.org/10.1021/acs.jctc.5b00935
-
Kucerka, N., Nieh, M.-P., Katsaras, J. (2011). Fluid phase lipid areas and bilayer thicknesses of commonly used phosphatidylcholines as a function of temperature. Biochimica et Biophysica Acta, 1808(11), 2761-2771. https://doi.org/10.1016/j.bbamem.2011.07.022
-
Vostrikov, V.V., Daily, A.E., Greathouse, D.V., Koeppe, R.E. (2010). Charged or aromatic anchor residue dependence of transmembrane peptide tilt. Journal of Biological Chemistry, 285(41), 31723-31730. https://doi.org/10.1074/jbc.M110.152470
-
Kovacs, F.A., Cross, T.A. (1997). Transmembrane four-helix bundle of influenza A M2 protein channel: structural implications from helix tilt and orientation. Biophysical Journal, 73(5), 2511-2517. https://doi.org/10.1016/S0006-3495(97)78279-1
-
Hanson, M.A. et al. (2008). A specific cholesterol binding site is established by the 2.8 Å structure of the human β2-adrenergic receptor. Structure, 16(7), 897-905. https://doi.org/10.1016/j.str.2008.05.001