Question:
I am having an issue with the eotest and FSC curves. When the particles are low pass filtered, the FSC curve produced by eotest starts to come up sometimes quite significantly, to around 0.5, at about the resolution frequency that the particles were filtered at. Also, the same thing happens when the models are CTF corrected. I have tested the effect of specifying different CTF correction cutoffs on the FSC curve at 20, 30, and 40A. Each time the curve produced by eotest starts to increase just before the cutoff. The same behavior is observed with the particles low passed using different low pass filters. My particles are negatively stained so I need to do a low pass filter for the reconstruction and the CTF correction is to a low resolution. Do you have any suggestions on how to eliminate this behavior?
Answer:
there are several possible causes for an FSC curve that rises in this way. 1. you may not be using a small enough ang= value for the size of map and resolution you are trying to achieve. This produces more local averaging, resulting in a slightly blurrier map, but, counterintuitively, actually makes the 'resolution' higher.
- An additional possibility is masking. This is particularly true for high contrast stain reconstructions. The FSC may contain correlations from the mask, not just the data. You need to make sure that your particles are edgenormalized (proc2d edgenorm) as the last step before reconstruction. You should also (once you have achieved a reasonably nice looking structure) use the xfiles= and amask= option, and insure that the last amask number is sufficiently large (about 10-15% of the box size).
- There may be some model bias issues, since the particles are not reclassified as part of the eotest procedure. On the other hand, reclassification is also not precisely fair, as it mixes model reliability with model resolution in the FSC curve, and these are really 2 independent items. When running eotest, make sure you are not using classiter=0, if 1 and 2 above don't cure your FSC problems, you can a) try increasing classiter back to something in the 5-8 range or b) you can run one or two cycles of completely independent reconstruction and compute the FSC between those. This is done by something like:
proc2d start.hed split.hed split=2 mkdir 0 mkdir 1 mv split.0.hed 0/start.hed mv split.0.img 0/start.img mv split.1.hed 1/start.hed mv split.1.img 1/start.img cp threed.?a.mrc 0/threed.0a.mrc cp threed.?a.mrc 1/threed.0a.mrc cd 0 refine 1 ... cd ../1 refine 1 ... cd .. proc3d 0/threed.0a.mrc 1/threed.0a.mrc fsc=psplit qplot peo psplit (to compare the old fsc with the new one)
If that STILL doesn't take care of the problem, ask me again :^)