Single cell Expression Analysis

Question

[David_emir]

Hi All,

I have 6,000 single-cell expression counts from 49 head and neck squamous cell carcinoma patients (20,000 Genes * 6000 cells /per sample). My question is how to conduct a differential expression of these samples, should I average (Merge) these samples to get a Raw count matrix or how to go about it? In general, how do you consider multiple samples in scRNA seq data analysis, since we consider cells as samples and Genes as features?

Thanks a lot, Dave

[Ram]

how do you consider multiple samples in scRNA seq data analysis, since we consider cells as Columns and Genes as Rows

Are the concepts of "samples" and "cells" not synonymous in single cell sequencing?

[David_emir]

I meant Multiple patients data in single-cell gene expression analysis. Should we treat each sample individually ? does that make sense? Sorry, I might be wrong, but somewhere I fail to understand this.

[igor]

To clarify, is it 300k cells total?

Answer 1

[jared.andrews07]

This is a very, very broad question. You can merge samples into a single matrix, but you must be aware of batch/patient technical variations that may just be noise in your data. These differences can be removed in a number of ways with a number of different tools, but discerning whether or not that's actually necessary or wanted is more difficult. In general, people will merge samples together, cluster cells after dimensionality reduction, and perform differential expression between the clusters.

I'm going to recommend you take a look at the Bioconductor book on single cell analysis, as it is the most comprehensive resource that I know of. It is, of course, bioconductor-centric, and there are many other options for most of the steps. But it's really nice to have it all in one place and does a great job explaining concepts/rationale behind various methods.

Answer 2

[swbarnes2]

I think you'd have to be more specific as to exactly what question you are asking, but I guess it wouldn't hurt to combine them all together just to see how well all the samples overlap, or don't overlap for starters., if your computer can handle that computation.