In my case, I found a significant difference between the results filtered to 6.0 and 6.5 A.
Using the cluster modes, I got 5 clusters of my target protein with at least 3 different conformations.
However, the density of the flexible region in the most dynamic conformation almost disappeared when the resolution was filtered to 6.5 A.
But it does exist when filtered to 6.0 A.
What does this mean?
Should I filter no more than 6.0 A ?
Why does it turn vague when filtered to lower resolution?
Just because there is noise around?
Should I filter to a higher resolution like 5.5 A ?
Also, if I want to see the conformational changes of glycans, is it reasonable to to filter to 10 A (about the size of a sugar molecule) or to higher resolution like 8 or 6 A?
BTW, the sum of the particles was used to run NU and had a resolution at 2.97 A.
And also, How can I determine how many modes to solve?
When set to 3 modes:
In my opinion, there is not really a correct answer to most of these questions as every dataset is different and can require different parameters to interrogate. It’s really a process of trial and error to find something that “works” and then you must interpret if it is “real,” which can be the hard part.
It is often the case (for me at least) that running a program with one set of parameters will find the interesting class while a similar run with slightly different parameters will not find the same result. As long as you get a discrete particle set in the end that refines to the volume you want to describe, then it is fine (but not absolutely necessary, the burden of proof required depends on what claim you want to make). Filter resolution and highpass are usually the parameters I work with to find what I know is there (from 3D classification). 3DVA seems fairly sensitive to “catching” motions that can be changed with the lowpass/highpass so changing between different filter resolutions can give very different results. I tend to believe something is real once a discrete particle set can be separated and refined and its volume corresponds to the 3DVA frames. For this you may refine the cluster mode particle output.
Perhaps your flexible region is more ordered in a rare class and you need a particular set of parameters to find that. Perhaps run 3DVA with filter resolution changed in 2 A increments and see how they differ.
Thank you so much for your helpful reply.
I have another question. If I’ve run 3DVA with 3/4/5 modes, and each looks the same when viewed as movies.
The first component (Mode) is almost the same. The second and third components in mode=3 patterns are similar but in opposite colors (red to blue/blue to red) with mode= 4.
How can I know which components were be used to reconstruct clusters in cluster mode?
I am quite confused if it’s unnecessary to run 5 modes.
Maybe 3 or 4 is enough because the variability of my protein is not that high?
Thank you for this response. Validating my own headache trying to parse the variability in my dataset in which I know there are at least 60A of movement in the primary component. Refining to noise is so easy and I’m still having trouble understanding the filter resolution and highpass resolution components I should use and how these are affected by “filter order”
I know the movements are real, but it’s like trying to find an unknown number of needles in a slippery haystack.
The 3DVA preprint and publication address this issue in the right half of Figure 1, noting that “when the object scale is similar to the motion scale, the linear model is a good approximation to motion [but] when the object scale is smaller than the motion scale, the linear model is less accurate.” Thus, you want to choose the resolution range based on the size the object and scale of the motion you want to model.
For example, representing the complete presence or absence of even small binding partners requires very low resolution components, and in that case the high-pass resolution should be > 300 Å (or not used). If there are large domain level changes, with negligible internal rearrangements, the low-pass resolution should probably be > 6 Å to suppress contributions from changing secondary structures.
Don’t worry about using the default filter orders. Just know that the high-pass filter is relatively sharp, and the low-pass filter is relatively soft. I.e. a low-pass of 6 Å will include significant signal in the 5-6 Å range.
Also, keep in mind that the resolutions for the 3DVA job and the 3DVA Display job are doing different things. The former is setting frequencies used in reconstructing the 3DVA components themselves, the latter is the frequencies used when generating calculating volumes (simple mode) or running particle reconstructions (intermediate and cluster modes). In this post I only speak about the 3DVA job.