skills/bioskills/bio-imaging-mass-cytometry-quality-metrics/scripts/run_qc.py

• cell_snr · function · L23-L29 — def cell_snr(counts): # two-component mixture on raw counts, reported as context: on toy Poisson data a GMM can # split pure noise into a spurious "positive" component, which is exactly why the # empty-channel comparison below -- not the raw SNR ratio -- is the decisive test
• signal_above_background · function · L31-L36 — def signal_above_background(counts, empty_counts, q=95, tol=2.0): # operational "did this antibody work" test: is the POSITIVE tail above the empty-channel # floor? A dim-but-real marker carries its signal in the tail; a failed channel's tail sits # at the empty floor. Comparing medians would wrongly fail a sparse-but-real marker whose # median is the negative floor -- exactly the dim != failed trap the skill warns about.