Is adding 1 to the nominator and denominator to avoid infinity as a value for fold change statistically sound? can these genes be called up or down-regulated? Or should these genes be referred to as "uniquely expressed in one condition" and no fold change applies to them?
From my side, it doesn't make sense to calculate the FD if the gene is not expressed in one condition. For example, many tissue-specific expressed genes only expressed in one tissue but not the others, so how could we interpret the FD of the gene between two tissues? What I usually do, is first convert the RNA-seq count data into normalized expression value (e.g. TPM or fpkm), and then filtered all low-expressed genes, and then find the DE genes. In your case, I agree that you should explain this as "genes uniquely expressed in one condition".
I would simply use a standard tool for differential expression (DE) analysis - such as DESeq2 or edgeR ( http://bioconductor.org/packages/release/bioc/html/DESeq2.html , http://bioconductor.org/packages/3.1/bioc/html/edgeR.html ). Both these take raw counts as input and do the transformation and normalization (incl adding a pseudocount) internally and repport a p-value and log2FC.
If they are significantly differentially expressed refer to them as that.
"Genes uniquely expressed in one condition" sounds like you will have to impose an arbitrary cutoff indicating what is expressed and not, which will always be less robust than the the DE way.
@ Kristoffer: I am using 3 different tools for DE: CuffDiff, DESeq2 and baySeq. I consider as DE only genes with significant p-values and log2FC>1. While DESeq2 is straightforward and gives me a log2FC value for the genes uniquely expressed in one condition only (by adding 1 to the nominator and denominator), CuffDiff gives no such value. So my question is really regarding the output from CuffDiff. I have to either include them in a separate list or manually calculate the log2FC then apply my threshold. Same thing applies to the output of baySeq as the tool doesn't calculate the log2FC altogether.
@Zineb CuffDiff do calculate log2 fold changes (look at the output file gene_exp.diff and iso_exp.diff). Btw CuffDiff adds a pseudocount in the order of ~0.0001 FPKM). With regards to baySeq if you will just have to do it yourself or rely on the other tools. I prefere a pseudocount of 1 (no matter the units) since 0 then is transformed to log2(0+1) = 0. It really does not matter that much as long as you write what you are doing.
Yes Kristoffer this is exactly what I do. I have a list of up- and down-regulated genes from the 3 pipelines. The question now is wether it's correct to include the genes with 0 read counts in one condition in the list of DE genes or keep them in a separate list for "uniquely expressed genes"?
As long as they are significant it is okay - and especially if you use a pseudocount of 1. Alternatively you can also just add two collumns to the table with the mean expression for each condition (as the CuffDiff output has) - then those cases will be easy to spot.
I advise you to use DESeq2 evaluating your differential expressed genes (DEG) because is more sensible than CuffDiff discriminating DEG in gene expression differential analysis. In fact DESeq2 as you noted DESeq2 solve your problem by adding 1 to each gene that reported 0 read as expression value.
Moreover, CuffDiff tool is useful for transcript or isoform and differentially splice gene analysis. However, several package are available for transcript analysis such as RSEM… and so one.
Reporting gene expression as fold change assumes that the gene is expressed in both samples/conditions, which will be wrong to say in case of uniquely expressed genes. I would prefer it to be reported as "uniquely expressed gene" along with a some information of its expression level with respect to some other known gene e.g. house keeping genes.
Ok if in your experiment house keeping genes are included. However, what I usually do for this kind of analysis is to set an expression value threshold based on the quality control analysis. Then differentially expressed genes as fold change mesurement will regard only expressed genes.
How are the values of RNA-seq Log2Fold change expressed in the publications?
The log 2 Fold change value can be positive or negative indicating up and down regulation of expressed and/or considered gene candidats. Several Pakage like DESeq, Edgr, RSEM, Cufdiff and so one allow to calculate that ratio value betwen i.e. pair of samples (control vs. case or case vs. control).
RNAseq data analysis?
I have RNAseq data comprised of 1 before treatment sample and 1after treatment sample, how can I get log2FC for each sample? Can anyone suggest please?
Hi,
fold change value represents the ratio between your treated sample and your control sample. You cannot calculate fold change for each sample.
Fold change value with regard detected expressed genes in transcriptomic survey give you an idea of that genes modulation (i.e. up regulated gene; if log2 FC >0 and/or down regulated if log2FC
Dear Dago, Thank you very much for your prompt reply. Indeed your suggestion will help me. Actually I have a doubt, should I do DE analysis with these pair, will it make any sense? Sorry for the silly question.... How do I know, whether the fold change value is comes from untreated or treated sample? (I mean Log2 FC) Because I want to correlate how gene expression changes affect the drug response? Thanks again!
Dear Barani,
I suggest you to read "Beginner’s guide to using the DESeq2 package" of Michael L. et al., 2014. This vignette contains guide line to process your samples for gene expression differential (DE) analysis.
Also, I think that you can perform DE analysis with your sample pair. However, the risk here, is that statistic supporting detected expressed gene modulation will feeble. You can try to set a stringent value with regard False Discovery Rate (FDR) in discriminating differential expressed gene with the purpose to contrast above mentioned sample pair concern.
Bye.
- Badges
- Science topic
- Similar topics
- Engineering
- Materials Engineering
- Powders