Different conditions in different platform ?

[ahy1221]

Dear developers, Congratulations and thank you for developing such excellent tool. I was wondering that would Augur still works if I have two conditions sequenced by different protocols? For example, assuming that cells in the healthy condition were all sequenced by 10X 3' protocol while cells in the disease condition were sequenced by CEL-Seq2, is that still possible to do the comparison by Augur? I noticed that the paper discussed something about technology difference between scRNAseq and STARmap, but I am not sure I really understand what this part actually saying.

Thank you very much!

Yao He

[skinnider]

Hi @ahy1221, the point of our scRNAseq vs. STARmap comparison was that very similar cell type prioritizations are achieved by Augur for the same tissue exposed to the same stimulus, despite the fact that these were profiled using very different technologies. We did not directly compare, e.g., unstimulated scRNA-seq cells to stimulated STARmap cells, and it is hard to imagine this is going to be optimal. Having said that, this kind of thing was one of the major applications of the first bespoke ‘batch correction’ methods for scRNA-seq (e.g., Figure 3 in Haghverti et al., Nat Biotechnol 2018) so it may be worth giving it a try after applying some kind of integration method. Extended Data Fig. 10 of our paper may also be worth a look.

[mchimenti]

Following up on this question, is Augur intended to be applied after integration of treatment and control datasets?

For example, if I have control cells and 3 drug treatments, should I integrate all four datasets first using Seurat's methods before applying Augur to the integrated object? I have attempted this and the results I get are AUCs very close to 1 for all clusters. Either every cell type is equally affected by the drug treatments (seems unlikely) or I am doing something wrong.

Any insight is appreciated. Thanks.

[jordansquair]

Hi @mchimenti, yes Augur should be applied after the integration and cell type annotation of your data (e.g., see Extended Data Fig. 10 from the original publication). The most likely reason you are seeing AUCs close to 1 is that the default assay of your Seurat object is set to 'integrated'. The genes selected by Seurat for integration should be used for clustering, but not exclusively for cell type prioritization. Can you check the result of DefaultAssay(object) ? If it is 'integrated', set it to 'RNA', or whatever your assay is called containing your main expression matrix. Once you set that properly re-run Augur. Alternatively, to avoid confusion around assays you can pull out your expression matrix manually, along with your meta-data, and input them manually as so:

input = GetAssayData(object, assay = 'RNA') meta = object@meta.data augur = calculate_auc(input, meta = meta)

[mchimenti]

Thank you, that seems to have improved the results. And thanks for providing this great tool, and for providing some nice visualization functions as well (I was thinking of ways to viz the results when I found you've already written them!)