Differential Gene Expression Analysis with low biological replicates

Question

[sp29]

I have two groups with 12 (6 + 6) samples that were generated from bulk-RNA-Seq experiments. While performing the differential expression analysis the adjusted-p-value came out to be 1 with both DESEQ2 (Wald, LRT) and EdgeR (Quasi-likelihood, LRT). After, skimming through a few posts and papers, I found that its the result of Benjamini & Hochberg (BH) procedure and low replicate counts per group. Can anyone suggest some other workflow or library that I should implement to extract differentially expressed genes from such low biological replicates?

[dsull]

Would just like to add: 6 replicates per condition is not a low number of biological replicates. I routinely use half as many replicates per group. It's not statistically advisable to go through a ton of different workflows until you get the results you want (i.e. a good number of differentially expressed genes). Also, I find it hard to believe that all your adjusted p-values are 1. Even on negative controls, DESeq2 and edgeR won't give adjusted p-values of 1 for all genes. (Edit: nevermind, this is possible for adjusted p-values as documented previously: same padj for all the genes after DEseq analysis ; not for raw p-values though). Follow ATpoint's suggestions so we can help you further.

[sp29]

I am quite new to RNA-Seq so I followed tutorials from bioconductor's website.For DESeq2 I have used DESeq(dds) and DESeq(dds, test="LRT", reduced = ~ 1). For EdgeR I have used glmQLFTest(fit,coef="Group2") and glmLRT(fit,coef=""Group2""). The plots are Heatmap with eucledian distance and PCA

Also, quasi and LRT tests gave me FDR as 1 for all genes. But wald test gave the same value (0.98) for all genes.

[dsull]

Based on the clustering from the plots, I can't really discern any difference between group 1 and group 2. If they were truly different, group 1 samples would mostly form one cluster while group 2 samples would mostly form another cluster.

So it's probably really the case that there are no differentially expressed genes.

Answer 1

[ATpoint]

The workflows are called DESeq2 ad edgeR, they are perfectly fine for want you aim to do, getting results from low replicate number. If the results are not significant, then it is what it is. For a more elaborate answer please post code and results of plotPCA, see the DESeq2 workflow for it, plus please some details what these samples are.