Metagenomic differential abundance analysis

Question

[imaparna27]

I performed taxonomic classification of de-novo assembled contigs using CENTRIFUGE and obtained counts of microbes,

                                      Sample1     Sample2
 Ruminiclostridium sp. KB18            100             21
 Ruminiclostridium sp. KB18.1           46             30
 Ruminiclostridium sp. KB18.2            2             89
 Ruminiclostridium sp. KB18.3           66              5

I performed following steps using DESeq2 for differential abundance analysis:

read.counts <- read.table("count.csv", sep=",",row.names=c(1), header=T) 
metadata <- read.table("metadata.csv", sep=",", row.names=c(1), header = TRUE) 
dds <- DESeqDataSetFromMatrix(countData=read.counts, colData=metadata, design
    = ~ condition) 
dds <- DESeq(dds) 
res <- results(dds, lfcThreshold=1, alpha=0.05, cooksCutoff=FALSE)`
summary(res)

After, applying alpha as 0.05 or 0.1, no differentially expressed microbes were found. Does this mean that none of the microbes are differentially expressed? Or There is some mistake in the DEG steps I'm performing?

[Asaf]

What is the structure of your data? How many samples for each condition? Also, what microbiome is it? Are the samples similar enough so there is a common ground for normalization? DESeq2 assumes most of the entities have similar expression (or abundance in this case) across all samples, does this assumption apply to your data?

[imaparna27]

Thanks for your response. There were 20 samples in my study 10 each for normal and diseased condition. Bacteria microbiome was used, however I couldn't understand what do you mean by "Are the samples similar enough so there is a common ground for normalization?". If any other approach is used for differential analysis in case of metagenome, please suggest.

[Asaf]

As I've mentioned, the assumption DESeq2 does is that most entities did not change. In microbiome research this is sometimes true (mostly in in-vitro systems) but usually is not. Other approaches like ANCOM are usually more appropriate.