Tutorial: Particle Picking Calibration
In cryoSPARC v2.13+, there is a new feature that directly calibrates pick scores against defocus, making it much easier to set thresholds when using the Inspect Picks job.
Particle picking on cryo-EM data can be a challenge due to the presence of various complicating factors such as junk particles, aggregations, ice crystal formation, carbon edges, etc. However, one of the simplest variances amongst images that is not accounted for by most particle picking methods is defocus variation within a dataset.
Defocus variation causes the same particles from the same viewing directions to look different, and also changes the signal and noise stastics of each micrograph. Thus, in particle picking tools, particles of equal quality from two micrographs with different defocus can be assigned different scores, making it difficult to set thresholds on picking scores in order to select good particles.
CryoSPARC's blob picker and template picker have been susceptible to this problem. In CryoSPARC v2.13+, there is a new feature that directly calibrates pick scores against defocus, making it much easier to set thresholds when using the
inspect picksjob type.
To use picking calibration, perform blob or template picking as normal. Then, connect the
particlesoutput to an
inspect picksjob, along with
micrographs. This job now has new parameters called
Calibrate Pick Score to CTFand
Calibrate Power Score to CTFthat are turned on by default. When you run the
Inspect Picksjob with these default parameters, you will see plots that show micrograph median pick scores versus defocus in the streamlog:
There will generally be a correlation for both the pick score and the power scores, which measure independently the shape and density of a particle candidate, respectively. You will also see fit calibration lines on these plots. After calibration, the scores of each particle will be recorded relative to the calibration line, and these values will be shown in the interactive part of the job that allows setting thresholds on the parameters.
With calibration on, you should notice that good particles are clustered together more tightly in the pick score vs. power score plot. You should also notice that setting a threshold to select good particles in a particular micrograph also yields good picks on other micrographs that have very different defocus: