search

f-hamidlab / nuclearpy

MIT License
0 stars 0 forks source link

Variable normalisation thresholds #16

Closed Marcel-Salier closed 2 years ago

Marcel-Salier commented 2 years ago

I opened the segmentations we did and I got this: image I couldn't find the right bins to get them aligned all. So I wonder how this violin plots corresponds to a real histogram. Maybe we have to filter before small dots that can be an issue if there are too many.

fursham-h commented 2 years ago

Which channel is this? How does it look like using median normalisation?

Marcel-Salier commented 2 years ago

Is DAPI. I tried several experiments at the same time. The weird thing is that the violin plots should show the mode but apparently, the mode end in a different place. It seems that the graphic is not representative of the calculations. If I look for the median it has similar unrecognisable shifts. Either the line or the violin plots are not right...

fursham-h commented 2 years ago

Nah, I've checked the plots and lines multiple times on different channels, they are correct. It's probably the calculation. Can you email me 2-3 of the segmented csv files @Marcel-Salier for me to troubleshoot?

fursham-h commented 2 years ago

@Marcel-Salier, I got it.

What you are seeing is the violin plot AFTER normalisation and the red line is the mode BEFORE normalisation. I've switched the plot to show the distribution BEFORE normalisation, if that's what you're looking for.

Marcel-Salier commented 2 years ago

Ok, the whole idea of having the red line is to visualize the mode and change the number of bins to something that makes sense since modes are very sensitive to that. Not sure why we end up with this mix...