Reproducing SC subtype calls

[miguecass]

Hi,

I am working with your data and I am trying to reproduce the SCPAM50 calls from the metadata provided for the ephitelial cells. I follow the code from Calculatingscoresandplotting.R and Highest_calls.R, but I am unable to replicate the results.

What I did was the following:

• I generated a Seurat object for each of the breast cancer subtypes from the samples used. Then I generated another seurat object for the test set, again using the samples as stated in the supplementary material. I normalized and FindVariableFeatures as stated in Calculatingscoresandplotting.R
• Then I find the anchors and I integrated the data.
• Here comes some complications for me. I checked the Github issue opened here, and the response here was that the integrated data between training and test was used for infering the SC subtypes. However, I see that in Calculatingscoresandplotting.R once we have integrated the data, there is a phrase saying: #Calculating SC50 scores on the 'RNA' scaled data, see here .Additionally in Highest_calls.R , we have the input Mydata (which I am not sure what it is) but from the code (here) we can see that : tocalc<-as.data.frame(Mydata@assays$RNA@scale.data). Meaning that it cannot be the integrated data. Then, I wonder, which is the data that is used to infer the SC subtype calls.

I would really appreciate if I could get some insight on this. I am not sure what the input files like "TNBCmerged_Training_SwarbrickInferCNV.txt" contain or even Mydata in here. What do those files contain exactly? (e.g: all cells or only ephitelial, are they curated cells, integrated or just normalized RNA data...). If I could get the input files used to generate those results it would help me a lot. And which is the script used, or the steps to follow to reach the calls available in broadinstitute.

Sorry in advance for the convoluted question.

Thanks in advance,

Best,

Miguel

[johnyaku]

This was before my time, but I will try to direct your inquiry to the people who conducted the analysis.

To help us get to the bottom of it more quickly, can you please edit your question to link to snippets in the scripts that you are referring to?

[miguecass]

This was before my time, but I will try to direct your inquiry to the people who conducted the analysis.

To help us get to the bottom of it more quickly, can you please edit your question to link to snippets in the scripts that you are referring to?

Hi, thanks for the reply. I edited my questions with the proper snippets. Hope it helps!

[jgtz]

I wanted to ask if there was any resolution since I am running into the same problem of not even being able to reproduce even the SCsubtype calls published in this data. Running Highest_calls.R with the data they deposited in the single cell portal gives subtype calls correlated with the ones in their metadata, but still very different from their reported calls.

Thanks!

Edit: link wrongly formatted
Jorge