Merging Biological replicates and confounders

[pandeyravi15]

Hi

I have 3 biological replicates from 10X genomic data. I want to do pool all of them and remove confounders like batch effect, cell cycle effect, nGene and nUMI. So I can merge three objects using merge Seurat. Then I am removing these 3 factors:

ScaleData(object = control.combined, vars.to.regress = c("batchid", "nUMI", "percent.mito"))

So, should I regress out cell cycle effect after this step or combined that in this step, but how? Also, where I can find S and G2M phase cell cycle genes for mouse to use here for cell cycle regression.

Thanks.

[mojaveazure]

Whether or not you should regress out cell cycle effects depends on your data. For actually regressing out cell cycle effects, please see #398

[leonfodoulian]

@mojaveazure,

It would be helpful to include in Seurat also a list of mouse cell cycle gene names. I have quickly created a list and stored it in .rds format (see mouse_cell_cycle_genes.rds), that you can find below.
mouse_cell_cycle_genes.zip

Included also is the function I used to convert human gene names to mouse gene names (see ConvertHumanGeneListToMM.R), as well as the script that I used to get these genes (see get_mouse_cell_cycle_genes.R).

I never regressed for cell cycle, so I am not sure to what extent my gene list is valid.

Hope this helps!

Best,
Leon

[pandeyravi15]

@mojaveazure
So, I need to first check for each datasets whether I have to regress cell cycle effects by PCA plot mentined in your cell-cycle regression vignette. If cell cycle effect is there, than in that case what would be best workflow?For e.g.,if I follow this:

1. merge all seurat object
2. scaled out batch effect, nGene, percent.mito,
3. cellscoring on scaled data
4. again scaled out cell cycle effect
5. downstream analysis

is that would be good?

Also, why only S and GEM phase cell cycle genes are considered?

Also list provided by @leonfodoulian is good enough?
@leonfodoulian
Thanks.

[pandeyravi15]

Cellcycleeffect_oldR1.pdf
If PCA plot for cell cycle genes looks like attached figure, that is mean I do not need to remove cell cycle effect, right?

[mojaveazure]

We consider S- and G2M-phase genes as these are the cycling genes. If a cell isn't expressing these genes, then the cell is likely in G1-phase, or not cycling. Furthermore, by scoring on these genes, we can regress out all cell cycle signal or only the cycling/uncycling status (see Alternate Workflow in the cell cycle vignette). As for whether you should regress out cell cycle effects, you can run ScaleData on your scored object and store it in a new object. That way, you can compare the PCA plots and use the object that produces the better plot.

[pandeyravi15]

Okay, so my question is when to do this step: before filtering out cell and scaling out for batch effect, percent.mito or after that?

[mojaveazure]

Ideally, at the same time. ScaleData scales and regress data stored in the data slot and stores the results in the scale.data slot. If you'd like to do this in separate steps, you would first scale on other effects, then pass that matrix to ScaleData in the data.use argument.

control.combined <- ScaleData(object = control.combined, vars.to.regress = c("batchid", "nUMI", "percent.mito"))
# Pass whatever gene list you're using, could be cc.genes$s.genes and cc.genes$g2m.genes
control.combined <- CellCycleScoring(object = control.combined, s.genes = s.genes, g2m.genes = g2m.genes)
# Could also use difference between scores. See Alternate Workflows of cell cycle scoring vignette
control.combined <- ScaleData(object = control.combined, data.use = control.combined@scale.data, vars.to.regress = c("S.Score", "G2M.Score"))

[pandeyravi15]

Okay, thanks.

Will you recommend to do just do homolog search for cell cycle genes for mouse as @leonfodoulian suggested or do you have genes specific for mouse

[mojaveazure]

We currently do not provide specific genes other than the ones in the cc.genes list. I'd suggest using @leonfodoulian's approach.

[vondoRishi]

@mojaveazure It would be nice if data.use = control.combined@scale.data can be highlighted in the Cell-cycle vignette itself. I have noticed this only after 3 months.
Thanks

[Sophia409]

@mojaveazure @vondoRishi In seuratV3.0, there is no data.use parameter in ScaleData.If I didn't use data.use = control.combined@scale.data in separate steps, does it has any bad effects? Or what should I do?

control.combined <- ScaleData(object = control.combined, vars.to.regress = c("batchid", "nUMI", "percent.mito"))
# Pass whatever gene list you're using, could be cc.genes$s.genes and cc.genes$g2m.genes
control.combined <- CellCycleScoring(object = control.combined, s.genes = s.genes, g2m.genes = g2m.genes)
# Could also use difference between scores. See Alternate Workflows of cell cycle scoring vignette
control.combined <- ScaleData(object = control.combined, vars.to.regress = c("S.Score", "G2M.Score"))

object	An object
...	Arguments passed to other methods
features	Vector of features names to scale/center. Default is all features
vars.to.regress	Variables to regress out (previously latent.vars in RegressOut). For example, nUMI, or percent.mito.
latent.data	Extra data to regress out, should be cells x latent data
model.use	Use a linear model or generalized linear model (poisson, negative binomial) for the regression. Options are 'linear' (default), 'poisson', and 'negbinom'
use.umi	Regress on UMI count data. Default is FALSE for linear modeling, but automatically set to TRUE if model.use is 'negbinom' or 'poisson'
do.scale	Whether to scale the data.
do.center	Whether to center the data.
scale.max	Max value to return for scaled data. The default is 10. Setting this can help reduce the effects of feautres that are only expressed in a very small number of cells. If regressing out latent variables and using a non-linear model, the default is 50.
block.size	Default size for number of feautres to scale at in a single computation. Increasing block.size may speed up calculations but at an additional memory cost.
min.cells.to.block	If object contains fewer than this number of cells, don't block for scaling calculations.
verbose	Displays a progress bar for scaling procedure
assay	Name of Assay to scale

[Sergio-ote]

Best answer to #2493 is the zip file in this thread