Batch Adjusted counts for Downstream Analysis

Question

[saicharanp18]

Hi All,

I need some help with removing batch effects. I have 21 samples 11 samples sequenced in batch A and 10 sample in Batch B. Both batches have both genotypes to compare.

1. So, In order to perform further downstream differential expression analysis, i need batch adjusted count matrix. How can i get this. i tried combat, limma removeBatcheffects() but it gives negative values for zero counts.

2. Can i use batch adjusted matrix as a count matrix and perform DEA.

Thanks in advance.

Answer 1

[segato.felipe]

Hello,

You could use combat to correct for batch effects, remove genes absent in most samples and perform your DE analysis using a statistical test. We did this in our paper: (https://www.sciencedirect.com/science/article/pii/S221112471931455X)

[ATpoint]

Just out of interest, why did you use such a custom DEG methodology including t-tests instead of any of the established tools such as limma/edgeR/DESeq2?

[saicharanp18]

Hi, Thanks for the reply. did you use combatseq or the original combat. if you used original combat how did you deal with the negative values in batch adjusted counts, because i had some problems with combat for rnaseq. so i switched newer version combat-seq which preserves count characteristics..

Best, sai

Answer 2

[ATpoint]

Typically one includes batch into the design such as ~ batch + condition. Check e.g. the DESeq2 and edgeR manuals for this. This would correct for the baseline differences induced by batch. Also please browse the web for this question, there are literally dozens of similar questions already at Bioconductor support forum and the developers of the standard DEG tools have extensively commented there.

[saicharanp18]

Thanks @ATpoint. I tried different DEG tools and adjusting batches in the design formula. Our RNASeq protocol is a little bit different than bulk RNA Seq so i had to do some outlier analysis for which i needed batch adjusted normalized counts. I ended up using newer version of combatseq, which preserves integer characteristic of count data.

Thanks, Sai

[rezaeir75]

I use this method with limma/voom. After getting differentially expressed genes, I'd like to make a heatmap from my count data just for the selected statistically significant DEGs. However, when I plot the heatmap, I still see that samples in each batch are clustered together instead of replicates being clustered together. So, how do you approach the downstream analysis in this method, because as I could understand this method considers the batch effect for DEG analysis but doesn't transform the data for downstream analysis. Should I use combat or removeBatchEffect after DEG analysis for my DEGs?

[ATpoint]

I gave my two cents at Bioinfo-SE: https://bioinformatics.stackexchange.com/questions/14157/removing-batch-effect-in-heatmaps-after-differential-gene-expression-analysis