[3dem] local 3DR and Class3D previous experience

[Carlos P. Mata]

Hi,

We are working with several datasets of an icosahedral virus (size  
400x400 pixels). After the complete image processing (Relion via  
Scipion), we obtained a 3DR at 3A resolution at the shell and worse  
(but not too bad) for the protruding spikes (4-5A resolution). To  
obtain a map of the spikes with improved resolution, we extracted the  
local areas corresponding to the spikes, following the procedure  
described by Ilca et al (Nat Commun. 2015, 6:8843) to generate a  
localized 3DR. After extracting the local areas, we used the  
previously determined orientations and Relion Class3D skipping the  
alignment to select “good” spikes. For one of the datasets this  
approach is successful and we are able to obtain higher resolution 3DR  
for the spikes using a subset of selected areas, so we are quite sure  
that we are extracting and applying properly the orientations for the  
localized reconstruction. However, for the other 2 datasets we do not  
get any good results, and ~95% of the particles go to one single class  
with generates a  much worse map than the original spike reconstructed  
with the capsid. Actually, if we just reconstruct the spike using the  
extracted areas and the already determined orientations, this 3DR is  
also really bad (very low resolution, ~30 Å) for this two datasets  
while for the other one is as good as the original one in the capsid  
context.

Datasets have been obtained from a Titan Krios microscope. The first  
dataset with a K2 camera while the other two with a Falcon III. We  
don't know if it is important, but when we look at the log from the  
Class3D, in the case of first dataset  the "CurrentResolution" is  
changing depending on the iteration; but in the case of other two,  
this value seems to stuck in a close to Nyquist value (2.16, our pixel  
size is 1.065 A/px). Other difference is that, in the two  
“problematic” datasets we distinguish the Fresnel ring at the base of  
the spikes/shell edge in the fullcapsid 3DR (~3A resolution). However  
in the “good” one (also 3A resolution) we do not detect any Fresnel  
ring at the capsid borders.

Anyone has previous experience with this kind of staff?

We are wondering if we can change any parameter to obtain results so  
good with the two last datasets as the ones we obtained for the first  
one… Any idea?

Thanks in advance,
Carlos P. Mata, PhD
Post-doctoral Researcher
Department of Structure of Macromolecules
National Center for Biotechnology/CSIC
Campus de Cantoblanco Darwin nº 3
28049 Madrid, Spain
Tel: +34 91585 5472
email: cperez at cnb.csic.es