I’ve been reviewing the guidelines on box setup, but I’m still unsure about something and would appreciate your input.
I’m simulating a nanodisc (~18 × 18 × 4.6 nm) in a 21 × 21 × 12 nm box using CHARMM36 and semiisotropic pressure coupling in GROMACS.
Cubic boxes are commonly used in tutorials, but in systems like mine — or for something like straight DNA (e.g., 40 nm in one direction) — the molecule is clearly not symmetric in all directions.
So I have a few questions:
Is it acceptable to use a rectangular box (e.g., 40 × 5 × 5 nm), as long as each part of the molecule has at least the minimum required buffer (e.g., 1.0–1.2 nm depending on the force field)?
Or should the entire molecule be centered with equal distance from all box faces?
If a biomolecule (DNA or peptide) gets too close to its own periodic image, will GROMACS detect and stop the simulation, or will it continue running with possibly incorrect interactions?
In my current nanodisc setup (disc height ~4.6 nm vs. box Z = 12 nm), would it be safer to switch to a cubic box (21 × 21 × 21 nm) to avoid artifacts?
Thanks in advance for your help and clarification!
I’m not a nanodisc expert, so mine are more open discussion points rather than straight advice. I guess checking for previously published studies would be the safer way for you to answer properly some of your doubts.
I would start by considering that the lipid nanodisc can rotate along the axis parallel to the membrane, so if it rotates vertically (supposing you start with a planar disc along XY) then it can interact with itself. In principle then I would say that you should use a box that is larger than the nanodisc size in all directions (so a ~20x20x20 nm) to be sure to avoid self interactions. But by symmetry then you can use some other shape of box probably? Like octahedron or dodecahedron, since you can rotate everywhere you have some kind of spherical symmetry, which will help you decrease the amount of water molecules in the box.
Regarding your points:
i) Yes but this should always be true also during the simulation. In fact, the -d flag that is used in gmx editconf to set the distances will take the distance wrt to every part of the molecule to all the sides of the box. This means that if you treat your nanodisc as a whole molecule with its own degrees of freedom, then having a distance of ~1.5nm on each side will produce a box that is ~20nm in length on each direction, because you can rotate, so the orientation you have at the beginning of the simulations is not necessarily maintained during the run and this might break the PBC conditions and induce artifacts. For large lipid bilayers this is not true anymore and you can have 20x20x10nm boxes for example, but this is because the xyz symmetry of the box is broken by having a continuous structure along xy that cannot rotate. This ensures that the box length along z doesn’t have to match the length of the bilayer along xy (but for very large boxes then lipids can have induce curvature so also here it’s not that trivial).
ii) GROMACS carries on. Notice that self interaction is not something that is numerically wrong in principle, so it’s not a code error but rather a model/physics error that has to be addressed by the users.
iii) For the points above, I would say yes. Also at this point for the same considerations I would also use a isotropic pressure coupling (c-rescale possibly).