Comparing untreated vs treated cells about cistopic

[cbravo93]

Hi @ccruizm !

You can create a combined cisTopicObject by merging the count matrices of your samples; for example, you can create the independent cisTopicObjects and then extract the matrices to combine from [email protected]; if needed, you can also extract the metadata from [email protected]. Then you can use the combined matrix as input for createcisTopicObject().

Hope this is useful!

C

[ccruizm]

Hello @cbravo93

Thanks for the suggestion. I have merged the matrices, as you suggested. It seems to work, but now I have a problem when I want to plot the tsne and heatmap:

Error in scales::hue_pal(l = 60:100): length(l) == 1 is not TRUE
Traceback:

1. plotFeatures(cisTopicObject, method = "tSNE", target = "cell", 
 .     topic_contr = NULL, colorBy = c("nCounts", "nAcc", "densityClust", 
 .         "orig.ident"), cex.legend = 0.8, factor.max = 0.75, dim = 2, 
 .     legend = TRUE, intervals = 10)
2. .plotFactor(coordinates, variable, feature.names, main = columnName, 
 .     dim = dim, colVars = colVars, plot_ly = plot_ly, legend = legend, 
 .     cex.legend = cex.legend, factor.min = factor.min, factor.max = factor.max, 
 .     cex.dot = cex.dot)
3. setNames(.distinctColorPalette(k = length(levels)), levels)
4. .distinctColorPalette(k = length(levels))
5. t(unique(col2rgb((scales::hue_pal(l = 60:100))(2000))))
6. unique(col2rgb((scales::hue_pal(l = 60:100))(2000)))
7. col2rgb((scales::hue_pal(l = 60:100))(2000))
8. as.character(col) %in% "0"
9. scales::hue_pal(l = 60:100)
10. stopifnot(length(l) == 1)

Error in scales::hue_pal(l = 60:100): length(l) == 1 is not TRUE
Traceback:

1. cellTopicHeatmap(cisTopicObject, method = "Probability", colorBy = c("densityClust"))
2. setNames(.distinctColorPalette(length(unique([email protected][, 
 .     variable]))), as.vector(sort(unique([email protected][, variable]))))
3. .distinctColorPalette(length(unique([email protected][, variable])))
4. t(unique(col2rgb((scales::hue_pal(l = 60:100))(2000))))
5. unique(col2rgb((scales::hue_pal(l = 60:100))(2000)))
6. col2rgb((scales::hue_pal(l = 60:100))(2000))
7. as.character(col) %in% "0"
8. scales::hue_pal(l = 60:100)
9. stopifnot(length(l) == 1)

It seems an issue with the predetermined scale on hue_pal. Do you know how I could solve this issue? Besides, I would like to use UMAP instead of tsne. Have you implemented it in your tool?

Thanks!

[JBreunig]

Would you mind sharing your code for merging the 2 - [email protected] and 2 [email protected] objects? I can try on my end to do the UMAP as well.

[ccruizm]

Sure @JBreunig

After creating an independent object for each sample, I extracted the count matrix:

matrix_ctr7d <-  [email protected]
matrix_ctr28d <-  [email protected]

Then using Seuratv3, I create a SeuratObject and merge the data, this to append the sampleID to each cell and later, one extract the raw count matrix:

seurat_ctr7d <- CreateSeuratObject(counts = matrix_ctr7d, )
seurat_ctr28d <- CreateSeuratObject(counts = matrix_ctr28d)
control.combined <- merge(seurat_ctr7d, y = seurat_ctr28d,  add.cell.ids = c("ctr7d", "ctr28d"))
control.raw.data <- GetAssayData(control.combined, slot = "counts")

With this combined matrix I create a new cisTopicObject:

cisTopicObject <- createcisTopicObject(control.raw.data, project.name = 'control')

So far, I have not merged the cell.data generated by cellranger into the object.

My concern its how the peaks are handled by cisTopic since the calling is performed separately on different objects, therefore the peaks do not overlap perfectly, and they might be treated as completely different features. I am having issues plotting tSNE to visually asses what is the contribution of each sample per cluster. I am not expecting a big shift or new clusters after the treatment.

I hope this might help you. I expect to solve the issue with the plotting as well so I can have a look at the handling of the merged matrix by cisTopic.

[cbravo93]

Hi @ccruizm !

• The hue_pal error should be solved now, it was related to an update in the scales package.
• There is a wrap function called runUmap to run Umap from a cisTopicObject (on the topic-cell, or the region-topic matrix, you will have to specify).
• For merging the matrices, it would be better if you used the same region set for all data sets before merging (e.g. do peak calling on the aggregate of all samples, or merge/intersect the peaks). If they are the same peak (with different start/end coordinates) they shouldn't be treated as a different feature.

Let me know if you have more questions!

C

[ccruizm]

Thank you very much for your help!