ST: clustering on each tissue section separately or clustering on all tissue sections combined in Anndata object

Question

[Ibrahim]

I'm novice in spatial transcriptomics and I've given an analysis task of merfish ST data, the data consists of different tissue section and I've been asked to focus on certain tissue sections but when I try to do PP and clustering on each tissue separately it gives clusters different than when I try to do PP and clustering on all the tissue sections, so do I have to work on each tissue section separately or work on the entire dataset and then see the observation on each tissue section?

Answer 1

[OmnibusX]

Based on my experience with analyzing combined data, I'd suggest starting by analyzing each tissue section separately.

This way, you can explore the cellular composition specific to each section. This is particularly important because differences between tissue sections can be more pronounced than the differences between cell populations within each section. As a result, clustering all sections together from the start may lead to the mixing of similar populations across sections, potentially masking smaller, yet significant, populations.

Once you have a detailed annotation for each section, you can then combine the sections for comparative analyses. This lets you compare cell populations across different tissue sections, using the detailed annotations you've already done. At this stage, batch correction methods can be applied for visualization purposes.

I hope my sharing can be of help

[Ibrahim]

Thanks for your help! I managed to find clusters specific to each section, which align with the literature. Now, I'm stuck because I can't find a Python package that processes and predicts spatial domains in sections. Specifically, I'm looking for a tool that can identify spatial domains in MERFISH data. How are spatial domains typically identified?

[OmnibusX]

Until now, in my lab, this task still heavily requires a pathologist. We also have not been successful in finding assistance from automatic tools.