Strange spatialShrunkenCentroids results when using imzML converted via Bruker's SCiLS Lab 2020

[Gordon Luu]

Hi all,

I've been comparing imzML files of the same dataset converted into imzML format using either flexImaging 5.1 or SCiLS Lab 2020. Running spatialShrunkenCentroid on flexImaging exported imzML files is fine, but files exported from SCiLS give this weird segmentation map that consists mostly of white pixels/empty space. Has anyone experience this and maybe has a fix or solution?

I've included images of the segmentation maps below and ion images to show that the issue only seems to occur after segmentation. None of these data are biologically relevant so the ion images/segmentation won't look good, I mostly care about what's going on technically with the segmentation.

Code: https://github.com/gtluu/timstof_cardinal_analysis

Dataset: https://drive.google.com/drive/folders/1wWgtDTIgTFtQtSxKjwbiNn2neODcFYoS?usp=sharing

flexImaging SSC segmentation map:

SCiLS SSC segmentation map:

flexImaging ion image:

SCiLS ion image:

Thanks!

- Gordon

[Gordon Luu]

Images don't seem to be attaching but can be found in the github repo or google drive link.

[Melanie Föll]

Hi Gordon,

apologies for the late reply and thank you so much for this superb error report!

Looking at your files it seems the Fleximaging file is "processed" imzML and the Scils file is "continuous" imzML file.

For continuous files resolution and units in the readMSIdata function are ignored, this would explain the different m/z and intensity axes. To bin the continuous Scils file to 2000 ppm you can use the 'mzBin' function. I assume that then the segmentation looks similar to the one from the Fleximaging file.

I generally trust the Fleximaging data more as the changes made to the data during Scils import (and export) are not transparent for me.

I hope that helps,

Best,

Melanie