I have dataset of scRNA-seq data measured at 10x for 4 patients and 4 healthy individuals. I would like to merge these eight scRNA-seq data and perform hdWGCNA. I know that there are two methods for merging scRNA-seq data: cellranger aggr and the IntegrateData function in Seurat R package. When performing hdWGCNA, which method should I use to merge the scRNA-seq data? I would be grateful if you could provide some guidance.
In single cell, merging datasets or samples is nothing more than adding cells to the features to cells count matrix, you can see it as a cbind in R where your cells are your columns and features (in scRNA-seq genes) are your rows. That been said, cellranger aggr by default will normalized your cell counts by (docs) :
average read depth per cell between groups before merging
That is an option, but you will lose the ability to check how this normalization step impact your dataset, especially because you have both patients and healthy individuals.
What I would advise you to do is to either run cellranger aggr with --normalize none and load your new aggregated matrix in Seurat, as you are using Seurat. Or, load your 16 10x runs one by one in Seurat and merge the objects.
From there you can normalize, scale, batch correct, integrate...
Thank you for your quick response. I now understand the function of cellranger aggr. I will try to merge dataset using Seurat. In that case, can I still use hdWGCNA?
hdWGCNA is done after clustering and cell type annotation so yes
Thank you very much!
According to the advice above, I have 8 h5-format files generated by CellRanger for 8 people's scRNA-seq data, so I first created a seurat object with the IntegrateData function in Seurat R package. However, I don't know what to do next. Should I then implement the IntegrateLayers or JoinLayers function before implementing hdWGCNA? I would appreciate your guidance. Or, if you have any pipelines or tutorials that show how to perform hdWGCNA on h5 files that contain multiple samples, please let me know.
Are the 8 samples from the same sequencing run ?
Before integrating your samples which will skew your raw counts, you can just merge your 8 samples with the merge function.
You continue with reducing your dimension, cluster your cells and project them (UMAP or tSNE).
When you project your cells you can color and split them by sample. If in all your 4 control sample overlap well with no cluster specific to one or two specific control sample, then you can proceed with your hdWGCNA analysis.
If your control are not overlapping at all you can either Integrate your sample, or use Harmony which is a bit softer on the integration.
Thanks for the advice. Sorry for the delay in my replying. This 8 samples are performted in one run of 2 samples each. In other words, 4 runs were performed. Is the method you suggested the best way to handle this situation?
Log in to answer this question.
