What is the best practice for scRNAseq analysis of multiple samples?

Question

[Tundup]

For example, I have datasets of two groups - healthy and disease

Total 10 samples in each group

Datasets mix from different studies (so there must be batch effects)

I want to run analysis

First, lets say soupX

Then Doublet Finder

Then seurat workflow with Harmony integration

Annotate Clusters

Subset subtype

At each levels, I have to run steps which requires fine tuning at sample level.

For example, resolution, dims, expected fraction of doublets (which needs to be set, based on recovered cell numbers, 10x documentation).

How should I do?

1. Should I specify such for each samples Or
2. Should I proceed with a function with fixed but okayish type value for all samples (like 0.7 resolution or 0.045 if i have 4000 to 10000 recovered cells). Is this a non-problematic way?

Although sample level fine tuning should be the best way to do, I want to know if group level (batch or condition) tuning is a good approach. Global level tuning is not suitable at all (one fixed function for all samples be it coming from any batch, condition, samples).

Thanks you!

Answer 1

[Bastien Hervé]

What I usually do is checking for ambiant RNA, doublets and QC (number of counts, features, mirtochondria level...) for each sample individually, as these steps are specific to sample manipulation.

Normalization, selection of features, scaling and PCA can also be done individually.

Then you can integrate your samples and run the clustering analysis on the integrated data. Check if your replicates overlap nicely on your UMAP/tSNE.

[Tundup]

Replicates overlap properly in both cases after integration.