How many replicates do you have? You need at least 3 biological replicates for Deseq2 and cuffdif, with more than 3 biological replicates EdgeR is recommended.

Hisat2/Star+Stringtie+Deseq2 use to works fine.

Maybe you find this useful for a fast insight of your RNAseq reads: https://www.rna-seqblog.com/statquest-pca-clearly-explained/

If samples coming from different batches, may be you can consider using RUVseq

https://bioconductor.org/packages/release/bioc/html/RUVSeq.html

Make RLE plots before and after normalization to check the bias and corrections.

Also, look carefully on your QC reports.

I wonder whether you trimmed low quality reads and removed adapters before the mapping step. If not, please have a look on Trimmomatic or Cutadapt tools. I also would check both QC reports and PCR duplicates in your experiment since it could lead to false DEGs.

I usually select genes based on adjusted p-values (lower than 0.05) and not by logFC, but there are several possiblities here, depending on your hypothesis/experiment/further validation steps.

Francisco Callejas-Hernández I have 6 biological replicates for each conditions.

Dear Simang,

a good way to test your pipelines is to download published data where you know what results have previously been reported. If you apply your pipeline and get similar results (different analysis tools lead to different results and even different versions will change the outcome), then you know your analysis works. There should be no need to specifically change parameters for brain samples. 6 biological replicates per condition is very good and should lead to reliable results.

If your pipeline works on other data but does not report any differential genes in your brain samples it might be due to one of the following issues:

- technical data quality is bad. Test with e.g. FastQC and possibly trim/filter the data

- biological data quality is bad - some samples are outliers/switched and ruin your statistics. I suggest to generate a principal component analysis plot to check how the samples behave. With six replicates you can identify outliers pretty well and omit them from the analysis. Please be sure not to "overfit" your data by omitting several samples that are slightly different from the rest

- there is no or only a slight change on biological level. Everything works, but your experimental setting does not lead to changes in gene expression (or you picked a time point where no change can be detected yet).

I hope this helps.

Best regards,

Alexander

Are you performing differential expression on gene level or transcript level? (And supposedly you are not including novel detected transcripts?)

In case you do the analysis on transcript level, and this is what you need, then I suggest you first do it on gene level to make sure you get signal from the data. You can afterwards think about how to do transcript level differential expression, which is more complicated and less reliable, especially since you have many expressed isoforms in brain. Note deseq2 should only be used for gene-level analysis.

As preliminary checks, do you obtain reasonable mapping rates to the genome and transcriptome? For mapping to transcriptome, you can use e.g. salmon. You can afterwards also use deseq2 on the salmon output.

And finally, is your treatment or condition producing a strong phenotype? Would you expect to see a big transcriptomic change? Or rather just a few differentially expressed genes?