Differential Expression Analysis in Single cell RNAseq

Question

[asumani]

Hi all,

I have scRNA-seq data prepared with Smart-seq2 protocol. I have been using HISAT-StringTie-Ballgown(Partea et al. 2016) for my bulk RNAseq data.

For scRNAseq I have aligned& assembled transcripts with HISAT-StringTie. Yet, I believe normalization and further processing of transripts is different for scRNA. Any suggestions for how to process that data?

Best wishes,

Asuman

[ATpoint]

Is this a model organism that you work with? I would not assemble transcripts from sparse single-cell data, there is little that you can actually gain given you have a reference transcriptome.

[asumani]

It is human B-cells. There are 88 samples subsetted from 973 cells sequenced in initial data(publicly available data). It is complex study design, I took only the cells(B cell isotype of IgE) I am interested in.

Could you explain a bit more of your comment?

[ATpoint]

Well, transcript assembly (like every assembly) requires good coverage to be reliable. Single-cell data are inherently sparse per cell, and if this is human for which excellent reference genome/transcriptome exists, then there is no reason imho to even bother with transcriptome assembly. I personally would use salmon to quantify the reads against a reference transcriptome, but you can also use something like featureCounts to get your count matrix. From there on you should follow guided tutorials, e.g. https://osca.bioconductor.org/ or the Seurat vignette (but I strongly encourage OSCA). So you have 88 cells now, what exactly do you want to compare?

[asumani]

I have no idea why my reply is not saved here.. You can find it below.

[asumani]

Thanks a lot!

I misused the word 'assembly', I am sorry. I have already 'merged' transcripts using StrinTie and created count matrix. For bulk RNA, I normally use Ballgown for differential expression analysis which is an R package used in third step of this estanblished pipeline. So my concern is if I should switch to scRNAseq specific tool for normalization/differential expression steps.

20 A-type, 68 B-type cells are present in total of 88 cells. I am looking at DE between these two groups. Hope you can give additonal tips regarding statistics/tools after that information.

[jared.andrews07]

I'd second the recommendation of the OSCA book @ATpoint linked, it has a solid explanation of differential expression in single cell data and how pseudobulk replicates can be utilized to reap the benefits both of the resolution of single-cell data and the robustness of bulk RNA-seq methods to compare between conditions (Chapter 14). But in this case, basic marker finding is probably all you want/need, but it also has a chapter on that (Chapter 11).