Handle edge case where both precision and recall are 0

[apsteinberg]

Hi there,

Thank you again for creating this wonderful resource. We've been deploying minda across hundreds of our samples, and we discovered one other small bug. Given that you compute precision, recall and the f1 score as:

precision = tp/(tp+fp) 
recall = tp/(tp+fn)
f1 = (2*precision*recall)/(precision+recall)

There is an edge case which can occur if a caller has tp = 0 in which f1 is undefined and minda exits with an error. I added a small if/else statement to handle this in the PR.

Please let me know if this makes sense.

Best wishes,
Asher

[mikolmogorov]

@asherbryant thoughts?

[apsteinberg]

Hi @mikolmogorov and @asherbryant,

Following up on issue #15. I made some updates to the fork of minda to incorporate breakend orientation. The major changes are:

• ALT field incorporates breakend orientation
• INFO field includes orientation as STRANDS subfield.

Minor changes I made are:

• INFO field incorporates CHR2 and END subfields to allow for bedpe conversion with svtools.
• bug fix for working with compressed vcfs.
• Tests for strand inference
• Fix for header line (I found that the placement of the commandline string in the header appears to yield errors when working with bcftools).

If these changes make sense upon review, it would be wonderful if they could be incorporated. Thank you again for building this great tool.

Best,
Asher

[mikolmogorov]

Thanks a lot Asher and @amcpherson! This looks super useful, @asherbryant will you please review? I can check on my end sometime next week. @aysegokce maybe you could give it a test too on our datasets?

[apsteinberg]

Sure thing! :) Thanks, we're looking forward to hearing what you guys think.