Missed infected cell - PlateK19rep1_20200506_095722_264 - A01-0004

[tischi]

{"plateName":"PlateK19rep1_20200506_095722_264","siteName":"A01-0004","pixelLocation":[1793.0054504318803,1215.5873244399177,0.0],"analysisVersion":"0ca4addd462c63ffb354d6c6f59814d309dd2127"}

@imagirom

I implemented the cell feature inspection.

This cell for example as a very high top30 value of 21487, but is not marked as infected.

Screenshot

[constantinpape]

@tischi Have you checked if this cell was marked as an outlier?

[tischi]

But in fact it was an outlier:

Cell 395; Table /tables/cell_segmentation_serum_IgA/serum_IgA_outliers; Feature is_outlier; Value 1

So that seems a big part of the issues/ confusions...

[constantinpape]

Can you also post the outlier_type?

[tischi]

Sure:

Cell 395; Table /tables/cell_segmentation_serum_IgA/serum_IgA_outliers; Feature outlier_type; Value too_large

[constantinpape]

Ok, that's what I thought. The upper size threshold of 10.000 pixels is too conservative. In any case, I think I will just deactivate most of this for the next run, and then we should determine the values from looking at the distribution over all the cells, not just from spot checking.

[tischi]

Same for IgG:

Cell 395; Table /tables/cell_segmentation_serum_IgG/serum_IgG_outliers; Feature outlier_type; Value too_large

[constantinpape]

Yes, in fact IgA / IgGuse the same segmentation (too_large` means too many pixels in the cell)

[tischi]

I think I will just deactivate most of this for the next run

Yes, I think there should not be too many outlier cells in general, and since we do everything median-based the cell QC seems to be currently doing more harm than good.