search

SydneyBioX / BIDCell

Biologically-informed deep learning for cell segmentation of subcelluar spatial transcriptomics data
Other
35 stars 5 forks source link

Suggestion: Use of predefined cell localisation instead of DAPI image #21

Open GBeattie opened 3 months ago

GBeattie commented 3 months ago

Hey,

Great looking package, like the approach!

It's possible I'm missing a more fundamental need for the DAPI image, but if it's only use is to assign transcripts to the nuclei this information is also included in the transcript files (at least from CosMx). If the need for a raw DAPI image could be bypassed in cases where transcript localisation has already been determined, this would streamline the package quite a bit and negate the need for downloading/handling burdensome raw images.

All the best, Gordon

xhelenfu commented 2 months ago

Hi Gordon, thanks for the suggestion. The nuclei masks are also used during the prediction step of the cell masks. So we work from the DAPI image to help ensure consistency of the nuclei through the pipeline