search

kharchenkolab / Baysor

Bayesian Segmentation of Spatial Transcriptomics Data
https://kharchenkolab.github.io/Baysor/
MIT License
152 stars 31 forks source link

Nuclear-expansion like approach #101

Closed maximilian-heeg closed 3 weeks ago

maximilian-heeg commented 1 year ago

Hi,

I was wondering if it is possible to tweak to Baysor parameters to get a nuclear expansion like approach. Basically, for our dataset, we have an extremely good DAPI segmentation, but only approximately 30% of the transcripts fall into the nucleus. However, running Baysor creates a lot more cells than expected, even if the prior-segmentation-confidence it set to 1. I have tried to change min_molecules_per_segment, but this did not seem to do the trick.

Would there be an easy way to do this?

Thanks again, Max

VPetukhov commented 1 month ago

Apologies for the late reply. I think the simplest way would be to remove all cells without nuclei if you think they aren't real. But if there is actually only 30% of transcripts in DAPI, there is a good chance that some DAPI aren't captured. So cells without DAPI could also be real. Are there evidence that DAPI expansion is a better option here? I can add it easily.

maximilian-heeg commented 1 month ago

Not really. The reason I was asking is, that Baysor always seemed to overestimate the number of cells in my samples. But with the new 10x multimodal segmentation and a high prior confidence, this got significantly better for me.