Deep EM hancer map resolution

Hi,

I did a refinement on the map I got from deep EM hancer and it showed me a worse resolution that the map before running deep EM hancer, though I expected it to be better.
How can I estimate the resolution of map from DeepEM hancer?

you did a refinement of the *particles, and perhaps used the DeepEM map as input. DeepEM is a post-processing operation - it does not calculate FSC, does not change anything about the particle orientations etc. It just makes maps (volumes) prettier and more polypeptide-like. The resolution you put in can be assumed to be the resolution you take out. Re-doing refinement (of the same particles that yield the volume that you submit to DeepEM) should have extremely small changes if any to the calculated FSC resolution. if it’s very different then I would expect some other issue.

Agree w/ CryoEM2, the resolution of your DeepEM map would be the same as the input, given it is just sharpening. Similar to how we typically look at the “deafult” sharpened map output from a refine and call it by the (masked) resolution of that refinement.

I would be wary of using DeepEM maps as volume inputs for any refine jobs, I would say there is little consensus in the field about the potential introduction of bias from the DeepEM model. If you incorporate this in the middle of processing all downstream results would have that bias (if you agree that there is bias). If you just DeepEM your final map, then you can just publish your half maps alongside and anyone who is skeptical of the bias issue could just re-sharpen w/ more traditional means if they care.

I would be wary of using DeepEM maps as volume inputs for any refine jobs, I would say there is little consensus in the field about the potential introduction of bias from the DeepEM model. If you incorporate this in the middle of processing all downstream results would have that bias (if you agree that there is bias).

This applies to any map that does not purely come from the dataset being processed, not only to the maps produced by DeepEMhancer. This is also why you should apply a strong enough low-pass filter to the reference map used for a refinement job. I think the default 20 Å low-pass filter for heterogeneous refinement is too optimistic, and tends to produce maps that look a lot like the input references. I found that 40 to 60 Å for the low-pass filter helps a bit.

1 Like

Depends on the object being imaged, but agreed. 20 Å can be OK for high symmetry, fairly featureless objects (apoferritin) but is nowhere near enough for others…

I’d never use a post-processed map for a reference (unless I’m desperate enough to use something from EMDB, in which case I’d dump resolution out to 50-60 Å…)