but is there a way in any program to include 2D templates as reference for 2D classification? I have particles which are very often not found as a class, which I have been lucky enough to see as an obviously correct 2D class with optimized parameters (16Å max res align namely). Since the class exists I know the view is in the data but missed in all other classification attempts. Would be great to give a starting template to improve the alignment. Fish for 2D classes from crazy heterogeneous data.
Hey @CryoEM1, @hgxy15 – FYI: we’ve opted not to support initial references in 2D class to avoid the potential for ‘Einstein from noise’-type hallucinations. We generally recommend avoiding the use of any strong priors in the early stages of processing.
@vperetroukhin Thanks for the reply ! From my understanding, we could always perform a reference-free 2D classification after fishing out the desired classes from a extremely “dirty” or heterogeneous dataset. Provision of the prior would only serve as the bait but not the final golden standard in this case.
Same here, I tried to hack the resume process of class2D_stream by manually substituting the reference, but the program seems to reconstruct all templates using the particles instead of using the connected class_average.
We manged to find another temporary workaround by dividing the whole datasets into very small splits and mixing each split with the particles that represent the desired classes selected from previous 2D attempts. Performing 2D on the resulting mixtures gave rise to reasonable 2D class averages in our case, but took incredibly long computation time given the large amount of splits.