by using “molmap” command, I form a mrc-file from alphafold3-structure in chimeraX.
I successfully “import 3D volumes” using default settings.
but when I try to create templates from it, cryosparc warned “ValueError: could not broadcast input array from shape (69,39,35) into shape (34,34,34)”
I notice some said this should be solved by using “volume viewer” or something in chimeraX, but can’t find specific instructions. can I have some suggestions?
You need a cubic volume (same number of pixels in each direction).
Just be very, very, VERY careful with doing this. There are extremely high risk to get search model bias (Einstein from noise).
2 Likes
sorry, I tried “molmap #1 20 cube 34” or “molmap #1 20 cube true” or “molmap #1 20 cube 34 true”
or “molmap #1 20 cube true 34”
but chimeraX kept saying expected a keyword 
May I know more about this command?
I’m kind of a beginner in chimeraX, do you have any learning videos suggestions for me?
Thanks a lot!
You need to download the latest daily, not the stable version. And I echo @daniel.s.d.larsson’s caution - be very careful doing this, it should not be necessary and is very easy to fool yourself, would not recommend unless you are 100% sure you know what you are doing
I would overlay and resample on a cubic volume (like any volume generated by cryosparc).
agree with extreme caution, but the saving grace is that you can (and should) run ab initio of any downstream particle set you identify to mitigate the chance of reference bias (though picking bias still exists - high chance you’re missing the depth of info in your dataset).
thanks a lot. I use this solution because my particle was basically “string”, maybe only little part of this has a firm structure.
I can see the string in high LP filter. but cryosparc can’t pick the right particle. one is beacause my particle is indeed not in a firm shape. the other one is beacuse I used lacey-carbon. the software always pick on the carbon instead of ice.
so I try to make particle-picking easier for the cryosparc by using template-picking.
Ab initio does not mitigate template bias of this sort - if you have introduced template bias by picking with projections of a structure (depending on what threshold you used for keeping picks), you can get a perfectly reasonable ab initio up to the resolution you filtered the templates at.
The key is to make sure you filter your templates aggressively enough that you can discern Einstein-from-noise classes from real ones.
If you have problem picking on carbon, I would recommend to use an AI picker e.g. Topaz or CrYOLO
thanks a lot, I’ll try
Hi, I also use “molmap” command for templet picking, and change the box size to a cubic by relion command as follows:
relion_image_handler --i input.mrc --new_box {box} --shift_com --o output.mrc
the output map have the same box size every direction, with the same pixel size as molmap
if I want to change the pixel size and remain a same box size, just run:
relion_image_handler --i {mrc_in} --rescale_angpix {angpix} --force_header_angpix {angpix} --o tmp.mrc
relion_image_handler --i tmp.mrc --new_box ${box} --shift_com --o output.mrc
then we get the mrc file with the pix size and box size we want.
But I wonder that if we use templet picker by AF3, how can we make sure to avoid Einstein-from-noise and how to adjust the risk? I have got very nice 2d classes, and the protein can be seen picked well in micrographs. Is there still be much noise mixture in the particles to decrease 3D reconstruction resolution?
after using af3-map, I’ll check particle to see if there’s einstein mistake. but for more acturate result and solution, sorry i also can’t be sure what to do
As mentioned previously in this thread, I would highly recommend steering clear of this approach.
3 Likes