Hi,

Sorry for the lengthy post and might be naive. I am pretty new to differential gene expression analysis and also to transcriptomics.

We did metatranscriptomics analysis of a mixed bacterial population in two different condition. Thre was quality issue with the extrected RNA samples. Later i ran deseq2 normalization (by first using estimateSizeFactors and then variance stablizing transformation (vst) transformation). After the full deseq2 analysis i can see huge difference in replicate of one sample (many of the genes are highly expressed in one replicate are not expressed in the other one). The other sample replicate is ok.

Is there any way i can tackle this issue without re-sequencing?

Secondly, as i am very new to deseq analysis, i am not sure if my steps are correct (though i followed http://bioconductor.org/packages/devel/bioc/vignettes/DESeq2/inst/doc/DESeq2.html and other awsome tutorials). Here is my process:

1) Feed raw count data of all the predicted genes within the sample (use featurecounts table). 2) Tranform the data to deseq2 compatible format and provide metadata 3) i have two samples (control vs treatment) in 2 replicates. i used treatment vs control for deseq2 DESIGN and constructed deseq2 object 4) removed zero count genes. 5) estimateSizeFactors and plot sample data (pcoa, Hierarchical to see differences in sample). 6) VST transformation and check tranformed data. 7) run DESeq 8) summarize results.

After this i created a dataframe df <- assay(varianceStabilizingTransformation(deseq object, blind=T)) and from this dataframe used subset of genes (genes whom i am intrested in) to generate heatmap of the results of these genes.

Is this a correct process?

On the DESeq2 side, I would usually recommend to look at the sample using standard QC metrics in a framework such as MultiQC. And if it seems like a failed sample across QC metrics, then remove and if necessary re-sequence.

That sounds like a typical DESeq2 analysis.