I am using blob autopick to pick large viral particles that have a pesky tendency to sit near the holes edge. I lost ~ 1/2 of my particles during extraction due to the automatic edge particle rejection. I’ve reduced my box size to try and rescue some, but the particles are large and thus require fairly large boxes.
Can the reject edge particles feature be switched off?
Follow up question; is it unwise to do so?!
I figure it prevents issues with alignment in classing and such. But i also figured if i could extract the particles i could use a fairly close mask during refinement. I have observed that often particles sit next to the edge without touching it so was hoping there was some way to extract them and see what i can get out of them.
The reject particles near edges feature cannot be turned off at the moment. In extraction, we count a particle as too close to the edge only if its extraction box protrudes outside of the micrograph, so there’s no actual data to extract. If we were to allow users to disable this edge rejection, we’d have to fill in the missing pixels with something, because CryoSPARC broadly requires square particles. We haven’t looked into this closely, but off the top of my head I don’t know of a strategy for filling in the missing pixels that I’d have a lot of theoretical or empirical confidence in (if anybody knows of any references in this regard, please chime in!).
That having been said, if you’re losing a large number of particles, it might well be the case that you’d get an overall better resolution with those additional particles, even if you have to aggressively mask and lose out on some of the delocalized CTF information. I’ll record this request and discuss it with the rest of the team.
Hi Harris, I think this would be beneficial for folks with especially large particles (e.g. viruses), where each particle can take up a substantial fraction of the field of view. In this case, the number lost to edge rejection can be significant, and I imagine symmetry could compensate somewhat for the loss of delocalized information on one side of the box.
Relion implements the classical approach from MATLAB’s padarray with ‘replicate’ - you get those streaks on the edge as the last values in each row or column are repeated out to the required size. It doesn’t create as many artifacts as a step edge. In practice the streaks usually end up outside the circle mask. If a particle is actually cut by the micrograph edge then I wouldn’t use it in any case.
If you want to try this behavior now, why not extract the particles in Relion and then bring them back into cryoSPARC?
Thanks so much for the quick reply
I realise I was somewhat mislead in what exactly edge rejection was doing, but this makes a whole lot more sense! I of course now see the challenge.
On your second point, I’d be curious about whether this is the case! It’s likely many of the 200,000 particles i lost are junk as the blob picker was definitely picking noise and ice. But i have to assume I’m losing some in tact particles. I figure with the symmetry they could still be of use (as olibclarke mentioned above).
Thanks for your time and would be keen to know if anything comes of discussion with the team James.
Thanks for the suggestion. I think i will try this if i can thanks I am not certain i can easily migrate between the platforms (working a remote server and uncertain if relion is installed) but this would be a great solution for us. our current particle estimates are a bit slim for my taste!
Absolutely agree here.
While I’m not a fan of selecting particles close to the edge anyway, sometimes with giant viruses (etc) you don’t have much of a choice. It can be the difference between selecting four particles on a micrograph and selecting zero. Of course, you can tighten the box up, but that comes with it’s own issues - that said, it’s often done anyway as having the box a little too tight can be all the difference between being able to reconstruct a 3D volume and running out of VRAM.
I’d also like to see some sort of warning added when re-extracting particles with a larger box which makes it much, much more explicit how many particles you’ve “lost” by increasing the box size by even a few pixels.
Thanks for the comments and suggestions – we will consider these when implementing this feature!