cellHarmony: combat on scRNA-Seq

[LUSITO ELEONORA]

Dear Altanlyze team, 

I’m running cellHarmony via Altanylze GUI. My dataset is composed by blood cells (Neutrophils) from two independent donors and bone marrow cells from other two independent donors. “My" data have been processed into two different times and hence I’m performing batch correction using combat outside cellHarmony Altanalyze. I’m correcting the two blood samples independently from the two BM samples. And now the point. The combat correction (that I performed after normalising the data) gives negative values due to the abundance of poorly expressed genes (quite normal in Neu). When I give in input the matrix to Altanlyze to process the reference the tool does not generate the folder ICGS-NMF with UMAP projection and clusters maybe because (I hope not to be wrong) of the presence of negative values in the matrix. But we absolutely need the clusters because we are interested in the identification of different states of the cells. The only two clusters that are generated are the ones of cells with genes having negative values versus cells having positive values. This is not our aim. How can I deal with this? 

Moreover following  the instructions (below) that I found on the on-line documentation how can I ask Altanalyze to run combat on preprocessed files (i.e. normalized by Seurat) using the command line? 

Thank you in advance and thanks a lot for maintaining such a great tool!

Best wishes

Eleonora

python AltAnalyze.py --platform RNASeq --species Hs --column_method hopach 
  --column_metric cosine --rho 0.2 --removeOutliers no --row_method hopach
  --SamplesDiffering 3 --restrictBy protein_coding --excludeCellCycle no
  --ChromiumSparseMatrix tests/demo_data/10X/input/filtered_feature_bc_matrix.h5 
  --expname cancer --ExpressionCutoff 1 --output /tests/demo_data/FASTQ/output/ 
  --FoldDiff 4 --runICGS yes

Rispetta l’ambiente: non stampare questa mail se non è necessario.
Respect the environment: print this email only if necessary.

[Nathan Salomonis]

Hi Eleonora,

Nice work capturing neutrophils with decent expression! Combat is a reasonable way to proceed, but I would initially actually try to co-analyze the two donors in AltAnalyze by merging the ExpressionInput/exp.* files (CPTT normalized). If you still see batch effects (after you get the ICGS-NMF heatmap you can include a groups file with the expression file to see which cells correspond to which groups), then the combat normalized analyses are better to proceed with. You can also you Seurat clusters for cellHarmony analysis using scripts we have on our website (happy to clarify these). 

As you note, we have seen that ICGS2 errors out with negative values. This can be addressed in an update to the software but what I have done manually when encountering this issue (and what we would do automatically), is subtract each value in each row by the minimum value in that row, which then proceeds to effectively run the analysis. I think cellHarmony will work with negative values, but can't recall offhand. The combat analysis is not run in an intelligent way in AltAnalyze right now and has really only been tested with bulk samples. The intelligent way is to have the common cell populations (post-cellHarmony alignment or joint analysis) denoted as "groups" for combat with the different batches specified, but I have not been successful using the python implementation of combat for this purpose (I have seen examples in paper but not reproduced their methods. 

Let me know if I can help further with these. Happy to!

Best,

Nathan

--
You received this message because you are subscribed to the Google Groups "Alternative Splicing and Functional Prediction" group.
To unsubscribe from this group and stop receiving emails from it, send an email to alt_predictio...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/alt_predictions/6861FC1E-CE9B-4ECF-9CA4-9E2B8B066222%40hsr.it.