I am still seeing what I assume are artefacts in the components generated by 3D-VA. It doesn’t happen for every dataset, but when it does it manifests as strong lateral banding in the components. In this case, the sample was a membrane protein in a nanodisc (260k particles, junk removed, consensus refinement to 3.3 Å). True heterogeneity is present in some of these modes, I think, but the transitions look a bit messy, I presume due to these artefacts.
I have tried applying a 20Å highpass filter, and increasing lambda all the way up to 0.5, but the artefacts remain in both cases. Suggestions much appreciated!
Update - I think this banding might be at least partially caused by the MSP protein that forms the nanodisc.
The mask that I was using had a very soft edge, which partially overlaps with the MSP density at the edge of the nanodisc. When I use a mask with an edge that falls off before reaching the MSP, the banding is much less prominent.
In general, is mask softness as important for 3D-VA as it is for refinement? Or can a mask with a narrower edge be used?
I take that back - still seeing the same lateral banding artefacts in another case where the mask definitely excludes the nanodisc.
In this case the first mode represents a genuine conformational transition, while I believe the others are artefactual.
I wonder in such a case whether it would be worthwhile having an option to perform clustering using a subset of modes, in order to exclude components that appear to be artefactual?
I’ve seen this too for proteins in detergents. This behavior is seen for only one of the modes, but not the others. I interpret this as observing the largest variations within this set of particles as being brought by movements in the detergent belt. For protein movements, I see them in the other modes. I haven’t seen much impact of the mask used for this particular behavior.
In the picture you attach, could it be that the nanodisc is vertical, and that you have lateral motions of the MSPs?
Hi vincent, yes that is a possibility - I did make the mask small enough that it should not be seeing the MSPs though