Is it possible to compare two star files in Cryosparc, which correspond to the same set of micrographs? To elaborate, consider a scenario where we have 1000 micrographs associated with a specific EMPIAR ID. Assuming we have a “ground_truth.star” file containing the actual particle coordinates, and a “predicted.star” file containing coordinates predicted by our algorithm, is it possible to compute metrics such as precision, recall, f1 score, and other relevant measures between these two star files?
While I understand that we can import the star files to extract particles and assess 3D resolution for comparison purposes, I’m seeking additional metrics that can help us evaluate the accuracy of predictions based on the ground truth star file.
I greatly appreciate your assistance. Thank you!
@admin @admins I am also looking for the similar comparison.
There are good libraries to bring your star files in as pandas DataFrames for your data science applications. One is
pyem.star another is
starfile from #teamtomo.
star1 = starfile.read("particles.star")
star2 = starfile.read("test.star")
x1 = star1['data_particles'][['rlnCoordinateX, 'rlnCoordinateY']]
x2 = star2['data_particles'][['rlnCoordinateX, 'rlnCoordinateY']]
pairwise = x1.dot(x2.T)
You will need to think about how to calculate useful measures with nonidentical particle sets, for example by considering each particle in the test set to be a true positive if its nearest neighbor in the ground truth is within a certain distance.
You can check out the duplicate particle removal code from
pyem.star for an example of doing this performantly (with a spatial tree).
@DanielAsarnow Thank you so much for providing meaningful insights.