Plates from new experiments, listed as having issues by Vibor

constantinpape commented 4 years ago

@metavibor posted this on slack, I am just putting it here so we can keep an overview and link this to issues for the individual plates:

[x] H1 well (no serum) ratio in some plates (K12, K13, K17 and K19) is looking weird = the ratio is high and by my current thresholds I would score it positive
[x] I need to repeat plates K20 and K21, these are overgrown ones and this is where my negative control (F1) has ratio above 1.3 and I would score it positive... if someone knows what happened there and thinks we can save the plate for analysis let me know
[x] plate 156 failed (this plate has low intensities in all channels but this was analysed before and gave results that made sense... with the new analysis it did not work https://github.com/hci-unihd/antibodies-analysis-issues/issues/67
[x] plate9_2rep1 failed, not sure why there is lots of "nan" results, I checked a few of "nan" wells and they look fine #63
[x] plate6rep2_wp failed (this is one of the white plates with high background, I guess the subtracted background value was not ok... ratios are messed up but Z scores are fine)
[x] one observation - on some of "K" plates (for example K12) the Z scores are really low across the entire plate, also for positive controls (column 1)... could this be a signal that infected vs control cells classification did not work properly? https://github.com/hci-unihd/antibodies-analysis-issues/issues/65
[x] the best looking data are in plates "311", "943" and "316", there we had Z scores up to 20. In all K plates there was never anything higher than 2-4 although I used the some of the same samples as controls between plates. What does this tell us? (see also https://github.com/hci-unihd/antibodies-analysis-issues/issues/65)

cc @metavibor @tischi @imagirom

tischi commented 4 years ago

all K plates there was never anything higher than 2-4 although I used the some of the same samples as controls between plates. What does this tell us?

To dig into this it would be very helpful to have the mad as a column in the table, because one reason for a low robust-z-score could be a high mad. @constantinpape I am not sure now, do we have the mad already in the table?

imagirom commented 4 years ago

@tischi I think that might not even be in the individual (per-serum-channel) tables. I will add it.

constantinpape commented 4 years ago

Ok, let me know once you added it, @imagirom; I will add it to the default table then and redo the tables

imagirom commented 4 years ago

@constantinpape The column was already in; it is called {channel}_{infected/control}_mad_of_cell_{sums/means}

tischi commented 4 years ago

Cool! I'd also need the xlsx tables with this... In order to do some plotting in R...

constantinpape commented 4 years ago

Cool! I'd also need the xlsx tables with this... In order to do some plotting in R...

Do you mean the per cell statistics tables we discussed this morning?

tischi commented 4 years ago

Also the cell tables for some other analysis, but in this case I meant the image tables with the mad included. If I can make your life easier by running some python script to produce these tables myself I am happy to do this, of course.

constantinpape commented 4 years ago

Ok, I am working on it right now, gonna let you know when it's there.

constantinpape commented 4 years ago

@tischi I restarted the computation for the following plates now:

  "/g/kreshuk/data/covid/covid-data-vibor/plateK12rep1_20200430_155932_313",
  "/g/kreshuk/data/covid/covid-data-vibor/PlateK19rep1_20200506_095722_264",
  "/g/kreshuk/data/covid/covid-data-vibor/20200417_203228_156",
  "/g/kreshuk/data/covid/covid-data-vibor/plate9rep1_20200430_144438_974",
  "/g/kreshuk/data/covid/covid-data-vibor/plate6rep2_wp_20200507_131032_010",
  "/g/kreshuk/data/covid/covid-data-vibor/20200417_132123_311",
  "/g/kreshuk/data/covid/covid-data-vibor/20200417_152052_943"

The results will be written to /g/kreshuk/data/covid/sandbox/for_tischi. The actual results should be the same as before, but I have implemented the changes to the tables you wanted. Also, for all these runs the well tables, image tables and cell tables will be exported as .xlsx files to the plate folder. Results should all be there by tomorrow morning.

constantinpape commented 4 years ago

@tischi @metavibor Re plate6rep2_wp: this is one of the plates where I have used the non-standard background: 'serum_IgG': 5000, 'serum_IgA': 7000. Maybe that has something to do with it?

tischi commented 4 years ago

I checked all above issues as done now, because the infected cell detection seemed to not have been working properly in this run in many plates.

constantinpape commented 4 years ago

I checked all above issues as done now, because the infected cell detection seemed to not have been working properly in this run in many plates.

Sorry, I don't quite understand what you want to say with this.

But in general, yes the infected cell detection needs to be improved, as I have said already.

tischi commented 4 years ago

There were checkboxes next to the issues, I thought to be checked when we addressed them. I think we addressed them all now. That's why I checked them. Maybe wrong?

constantinpape commented 4 years ago

There were checkboxes next to the issues, I thought to be checked when we addressed them. I think we addressed them all now. That's why I checked them. Maybe wrong?

No, that's alright. I just couldn't parse the sentence you wrote before ;).

hci-unihd / antibodies-analysis-issues

Plates from new experiments, listed as having issues by Vibor #64