Dear all,
what is the difference between “Filter resolution” and “High pass resolution” parameters in 3D classification job, and how the optional last parameter may affect 3D classification if it has higher or lower value than set in “Filter resolution”?
Thank you!
1 Like
Hi @andro!
The high pass resolution parameter describes the frequency below which all signal is ignored. For example, if you set this parameter to 40 Å, then only differences in frequencies above 40 Å are considered. We often hear people use this when there are large, unstructured regions which are not interesting (for example, a micelle).
The filter resolution is a lowpass filter applied before classification. Essentially, it is the opposite of the highpass filter: only resolution below this parameter is considered. This is a very important parameter to the performance of 3D Classification. In general, it should be set to the lowest resolution at which you can still see differences between the classes. I like to use Volume Tools to lowpass filter my consensus refinement to various resolutions and find the lowest resolution where I can still distinguish the domain I’m classifying on from the rest of the target.
I hope that clears up these two parameters!
1 Like
Hi rposert!
Thank you for clarification!
Maybe to name these parameters as low-pass and high-pass filters, to avoid confusion, especially for new users?
Sorry, the next sentence is not clear for me “We often hear people use this when there are large, unstructured regions which are not interesting (for example, a micelle).”
If disordered regions are not interesting, what is the reason to apply High pass filter, which leaves only high (edited) frequency signal?
You’ve got it inverted - the high pass allows high frequency signals through, the low pass allows low frequency signals through.
Because it’s disordered, a micelle only has appreciable signal at low frequencies, this will dominate alignments when traditional cryo-EM safety mechanisms are employed, and can result in poor final alignments if the refinement moves to local refinement while on an “optimum” alignment where the protein is badly aligned compared to the micelle. It’s why for small membrane proteins refinements are often started at higher than classical resolutions (e.g. 12 Ang rather than 30 Ang)…
As an aside, disordered regions can be interesting (e.g. intrinsically disordered proteins (IDP)) but they’re a nightmare to work with due to their… well, disorder. 
5 Likes
Hi rbs-sci!
thank you for explanation and the note about the frequency. I corrected this.