Question: Batch effect correction in DE analysis of single cell RNA-seq data and visualization
0
gravatar for Poorya Parvizi
9 months ago by
The University of Edinburgh
Poorya Parvizi40 wrote:

Hi Everyone,

In my experiment, I have cells (same cell type) with 5 different drug treatments. These cells are sequenced in 3 batches and the goal is to find differentially expressed genes between treatments.

  1. I use SCDE for the differential expression analysis. SCDE accept gene read counts as an optimal input and there is an argument in the script that it takes batch information to deal with batch effect (In the PCA I can see the batches). In the end, I have differentially expressed genes and their significance, but it doesn't give an access to the user its batch effect corrected values (I guess this is also true in Seurat).

  2. As I can't get corrected values from SCDE and in order to continue the analysis based on DE genes, I used the Limma removeBatchEffect function on my read counts (or TPM or log2(TPM+1)). In this case, In outputs, I get negative values (probably due to regressing out) or all zeros in each batch change to the slightly bigger or smaller values (i.e. 0 to 0.1234).

What is the solution to get proper batch affected corrected values to continue the analysis with? Is the thing I do fine? My batch effect correction methods in SCDE and Limma are different, Is it okay? What people usually do.

ADD COMMENTlink modified 9 months ago by Praneet Chaturvedi120 • written 9 months ago by Poorya Parvizi40
2
gravatar for Praneet Chaturvedi
9 months ago by
Cincinnati Children's Hospital and Medical Center
Praneet Chaturvedi120 wrote:

Please use SVA package in R : https://bioconductor.org/packages/release/bioc/manuals/sva/man/sva.pdf

SVA can help you with identifying surrogate variables in the data and also account for batch effect removal using COMBAT

SVA is easy to use and if you find no surrogate variables then just use betweenLaneNormalization from EDASeq package with upper quantile normalization.

Also, you can try RUVSeq [https://www.bioconductor.org/packages/devel/bioc/vignettes/RUVSeq/inst/doc/RUVSeq.pdf] with RUVg normalization where you take counts of ~5000 least DE transcripts to normalize the whole counts matrix. This normalization is very effective with batch effect and sequencing effects.

Cheers !!

ADD COMMENTlink written 9 months ago by Praneet Chaturvedi120
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1045 users visited in the last hour