3D Flex solvent mask creation

Should 3D flex solvent masks exclude a micelle or similar, as we do for 3DVA masks?

1 Like

Hi @DanielAsarnow,

Good questions. We’re not quite sure what the best strategies are in all cases yet :slight_smile:
Since there is often a lot of variability in the micelle region, it makes sense to try and instruct 3DFlex to ignore that variability (unless that is of interest).

There are a few cases.

  1. The transmembrane region is “rigid” and carries a bulk of the mass of the protein, and so the micelle deforms but doesn’t move in an overall position sense. The flexible/moving parts are smaller soluble regions that extend from the TM region. This might be typical for e.g. ion channels like TRPV1 that we showed an example result for. Here, you don’t really care that 3DFlex models the micelle at all. So it might be reasonable to exclude it from the solvent mask entirely. There can be pitfalls with this though - keep in mind that 3DFlex is not constrained in terms of moving density outside of the solvent mask. The solvent mask only controls where density can be located in the canonical map. We found that in some cases if for example the mesh is fine enough and the rigidity is not high then when the micelle is completely excluded, 3DFlex will just learn to add extra density in the periphery of the TM region and then move that density outward, outside the solvent mask, to “reform” the micelle (since it is present in the data so 3DFlex wants to explain it). This is cool but not good!
    On the other hand, including the micelle in the solvent mask here will mean that 3DFlex can focus on the size/shape changes of the micelle.
  2. The TM region is a smaller part and most of the mass/rigid regions are soluble regions. Here, the TM region may move relative to the rigid parts, and the micelle therefore moves along with it (in addition to deforming). E.g., GPCRs, Rotary ATPase, etc.
    Here, the position of the micelle itself moves, so becomes more tricky to deal with. It should be included in the solvent mask so that 3DFlex can move it around.

In both cases a reasonable approach that we’ve tried but not explored completely is to use the segmentation options in the Mesh Prep job to treat the micelle region separately, and have it be part of the solvent mask but make it “rigid” so that it can still move around as a rigid body (in case the TM region is moving overall) but 3DFlex is discouraged from modelling deformations of the micelle itself.
One thing to watch out for with the segmentation is if you do create the micelle as one segment and the protein density as another, and then connect them together using the connections list, then the sub-meshes for the two parts will be fused. Tetra cells for the micelle will be considered rigid, and tetra cells for the protein will be non-rigid. But, because of the mesh fusion, nearby soluble parts of the protein that are close to the micelle (but not TM regions) will end up being fused to the rigid micelle and therefore unable to move. So for this, you should cut the protein into parts as well, so that you have micelle, TM region inside micelle, and soluble protein parts. Then connect Micelle > TM, TM > soluble parts. This way, the soluble parts won’t be fused to the micelle mesh and will be able to move freely.

We’re hoping to find better ways to deal with micelles (and mesh creation generally) - if you have any ideas or try anything that’s interesting please let us know!


Hi @DanielAsarnow @apunjani ,

for what it’s worth, we don’t exclude the detergent belt (it’s technically speaking not a micelle), instead we dilate the mask by a large number, say 20-60Å from the output mask of NU refinement, and we feel it gives us better results in defining the movements.
We have investigated many mask sizes, and the movements seen are always the same which ever size of the mask we’ve used. With larger masks, we observe the movements are better resolved than with smaller masks. Sometimes, we observe that with varying mask size, the order of the principal component might change (but not always), but still defining the same movements.
Also, while we were seeing detergent belt movements in the early components in cryosparc v2, in v3.3 we don’t observe them anymore. I’m in the process of upgrading to v4 so haven’t tested it yet.
Our protein is a 130kDa membrane protein, plus detergent belt of 150 DDM and 30 Cholate (quantified), which adds up a total mass of 220kDa.


1 Like

@vincent thanks for the detailed info! I think this will be very helpful for others working on membrane proteins with 3DVA (and 3DFlex).