Checklist for new results

hci-unihd / antibodies-analysis-issues

Issue tracker for problems in the antibodies analysis workflow.

0 stars 0 forks source link

Checklist for new results #71

Open constantinpape opened 4 years ago

constantinpape commented 4 years ago

Things to check for in the new results:

[x] Does the infected cell detection work properly now? (Cell outlier detection has been deactivated).
[x] Compare sum and mean based scores. (Sum / mean are computed correctly = with / without nuclei now)
[x] How does the new barrel correction look now.
[x] Check the plate plate5rep3_20200507_113530_429; something was weird with that one, feature computation took 3 times longer than for the other plates.
[ ] For future runs: can we estimate better parameters for the QC

cc @tischi @metavibor @imagirom

tischi commented 4 years ago

[x] For the sum scores in several plates the sums for the control cells are on average higher than for the infected. We need to figure out why that is (IgG_diff_sums = IgG_infected_q0.5_sums - IgG_control_q0.5_sums) => https://github.com/hci-unihd/antibodies-analysis-issues/issues/72

   plate_name                        `median(IgG_diff_sums, na.rm = T)`
   <chr>                                                          <dbl>
 1 20200417_132123_311                                           72684.
 2 20200417_152052_943                                          108659.
 3 20200417_203228_156                                         -221419.
 4 20200420_152417_316                                           38087.
 5 20200420_164920_764                                          -20008.
 6 plate1rep3_20200505_100837_821                              -230994.
 7 plate2rep3_20200507_094942_519                               304448.
 8 plate5rep3_20200507_113530_429                               -93284.
 9 plate6rep2_wp_20200507_131032_010                            169881.
10 plate7rep1_20200426_103425_693                              1329332.
11 plate8rep1_20200425_162127_242                               835289 
12 plate8rep2_20200502_182438_996                               815930.
13 plate9_2rep1_20200506_163349_413                             234479.
14 plate9rep1_20200430_144438_974                              -139018.
15 titration_plate_20200403_154849                             1543817.

tischi commented 4 years ago

new barrel correction looks now much better! 👏 @imagirom

did you compute for each plate it's own correction or did you average over all plates?
could maybe still be a bit better, upper left and lower left are a bit darker, around 50-100 gray values (I think in all channels). Here I blurred an image in an H01 well to make it visible:

imagirom commented 4 years ago

@tischi Nice!

did you compute for each plate it's own correction or did you average over all plates?

I averaged over all the plates, to reduce the local variations. We could do it for every plate individually, but that would not work well if some new plate has only few wells filled.

could maybe still be a bit better, upper left and lower left are a bit darker, around 50-100 gray values (I think in all channels). Here I blurred an image in an H01 well to make it visible:

If that is the case for all plates, it is a bug, since with the current method the average over all images should be a totally flat image. It might also have to do with there being less cells on that side of the plate on average.

constantinpape commented 4 years ago

@tischi thanks for checking all things. What's left to do is estimating better QC parameter ... I will export fixed cell tables for this soon.