Open ElyasMo opened 3 years ago
Hi,
I cannot comment on your specific results without actually seeing them and understanding why you think sctransform normalization does not lead "to a correct clustering of the spots".
But I notice that you run FindVariableFeatures
after SCTransform
. That step is not necessary (and might be a disadvantage) since SCTransform
already finds and sets the variable features.
I met the very simmilar problem with you. My dataset have many doublets, and the heterotype doublets formed a long 'bridge' between clusters. The sctransform pipeline cannot tell 'bridge' from normal clusters, but the seurat intrinstic 'Normalize' worked well.
Dear @ElyasMo
Why didn't you use SCT integration?
Dear ChristophH, Do you have a tip when we should not proceed with SCTransform. I have several spatial transcriptomics dataset and it has been a few weeks that I am trying to normalize my data through SCTransform. I just figured out that this method does not lead to a correct clustering of the spots while simple normalization works much better after regressing out the ratio of mitochondrial genes. So, I was wondering why SCTransform does not work well for my data while it is a powerful and highly recommended method of normalization.
One more thing is that, I have followed exactly the same workflow and codes in windows and mac and, surprisingly, I received slightly different clusters. How is it possible?
I am just adding my codes which is based on the standard tutorials and I have not added any new things.
Thank you very much in advance.
Here is the information from my datasets after pre-processing/filtration:
And here is my sessionInfo: