Question: How do I explain the difference between edgeR, LIMMA, DESeq etc. to experimental Biologist/non-bioinformatician
gravatar for Mike
21 days ago by
Mike880 wrote:

Hi All,

Could you please help me how to explain different methods for differential expression analysis such as edgeR, Limma, DESeq etc to biologist or non-bioinformatician.

Thanks in advance!

edger limma gene • 285 views
ADD COMMENTlink modified 21 days ago by Kevin Blighe9.0k • written 21 days ago by Mike880

What sort of difference do you really want to explain here? For the most part, just saying, "these are various tools for doing the same basic thing" is sufficient. That's then exactly the same as what happens in the wet-lab, where there are different vendors for similar antibodies/kits/etc. and they all give similar but slightly different results.

ADD REPLYlink written 21 days ago by Devon Ryan73k

Thanks Devon. I want to explain them some basic statistical advantage of these methods.

ADD REPLYlink written 21 days ago by Mike880

You could probably go into a bit of basic detail (like "edgeR uses negative binomial distributions which is good for ...") in explaining the statistical advantages of each of them.

ADD REPLYlink written 21 days ago by Hussain Ather510

Comparing them to each other? There's no reason to explain that to a non-bioinformatician.

ADD REPLYlink written 21 days ago by Devon Ryan73k

Not sure I could explain the difference between DESeq and edgeR to a bioinformatician TBH

ADD REPLYlink written 21 days ago by russhh2.8k
gravatar for Kevin Blighe
21 days ago by
Kevin Blighe9.0k
Kevin Blighe9.0k wrote:

DESeq and EdgeR are very similar and both assume that no genes are differentially expressed. DEseq uses a 'geometric' normalisation strategy, whereas EdgeR is a log-based method. Both normalise data initially via the calculation of size and normalisation factors, respectively.

Limma is different in that it nrmalises via the very successful (for microarrays) quantile nomalisation, where an attempt is made to match distributions across samples in your dataset. It can somewhat loosely be viewed as scaling each sample's values to be between the min and max values (across all samples). Thus, the final distributions will be similar.

Here is further information (important parts bolded):


DESeq: This normalization method [14] is included in the DESeq Bioconductor package (version 1.6.0) [14] and is based on the hypothesis that most genes are not DE. A DESeq scaling factor for a given lane is computed as the median of the ratio, for each gene, of its read count over its geometric mean across all lanes. The underlying idea is that non-DE genes should have similar read counts across samples, leading to a ratio of 1. Assuming most genes are not DE, the median of this ratio for the lane provides an estimate of the correction factor that should be applied to all read counts of this lane to fulfill the hypothesis. By calling the estimateSizeFactors() and sizeFactors() functions in the DESeq Bioconductor package, this factor is computed for each lane, and raw read counts are divided by the factor associated with their sequencing lane.



Trimmed Mean of M-values (TMM): This normalization method [17] is implemented in the edgeR Bioconductor package (version 2.4.0). It is also based on the hypothesis that most genes are not DE. The TMM factor is computed for each lane, with one lane being considered as a reference sample and the others as test samples. For each test sample, TMM is computed as the weighted mean of log ratios between this test and the reference, after exclusion of the most expressed genes and the genes with the largest log ratios. According to the hypothesis of low DE, this TMM should be close to 1. If it is not, its value provides an estimate of the correction factor that must be applied to the library sizes (and not the raw counts) in order to fulfill the hypothesis. The calcNormFactors() function in the edgeR Bioconductor package provides these scaling factors. To obtain normalized read counts, these normalization factors are re-scaled by the mean of the normalized library sizes. Normalized read counts are obtained by dividing raw read counts by these re-scaled normalization factors.



Quantile (Q): First proposed in the context of microarray data, this normalization method consists in matching distributions of gene counts across lanes [22, 23]. It is implemented in the Bioconductor package limma [31] by calling the normalizeQuantiles() function.


ADD COMMENTlink written 21 days ago by Kevin Blighe9.0k

Thank you Kevin for explanation.

ADD REPLYlink written 20 days ago by Mike880
gravatar for Istvan Albert
21 days ago by
Istvan Albert ♦♦ 75k
University Park, USA
Istvan Albert ♦♦ 75k wrote:

The simple explanation is that no statistical modeling can fully capture biological phenomena. Statistical methods all rely on assumptions and requirements that are only partially satisfied.

Different methods rely on different assumptions, and in turn, may capture different projections of the reality.

The different results are not mutually exclusive. One of them, neither or all may be true - all at the same time! Which is a bit hard to wrap your head around.

The true problem with all statistics is the "fakeness" and never-ending misrepresentation of the "p-values". P-values are not the accurate quantification of reality.

ADD COMMENTlink modified 21 days ago • written 21 days ago by Istvan Albert ♦♦ 75k

My opinion is that more emphasis should be placed on how well does a model reasonably capture reality, than necessarily just P-value as a scapegoat. The p-value is just an end product of many steps (see ). The same modeling and processing problems impact Bayesian statistics at least as much. Although you could argue much more attention should be paid to effect size.

ADD REPLYlink written 21 days ago by Collin450

The reason for scapegoating is that as long as p-values are permitted to be used, many/most don't feel the need to look for an alternative.

ADD REPLYlink written 21 days ago by Istvan Albert ♦♦ 75k

There certainly is an over-reliance on P value, and what's worse is that many don't know which test to use given the data distribution at hand.

My latest rule of thumb is to always do some sort of 'cross validation' analysis, i.e., re-do analyses multiple times and change parameters slightly. The genuine biological differences should more or less come out each time, whilst the questionable stuff will come and go from the results based on minor parameter configuration.

Unfortunately, we have already permitted sub-standard technology to enter research labs and now we have to deal with the high level of noise they bring.

ADD REPLYlink written 21 days ago by Kevin Blighe9.0k

I'm a proponent for cross-validation, especially ones with carefully constructed biologically meaningful folds. Take mutation effect prediction for example. A reasonable thing would be to group all mutations within a single gene into the same fold. That way the generalizability across genes is actually being accurately measured.

ADD REPLYlink written 21 days ago by Collin450

Thank you Istvan for your help.

ADD REPLYlink modified 20 days ago • written 20 days ago by Mike880

p-values are indeed hard for many to interpret. Also, for many genomics datasets, the point null of no change is trivial to reject. We try to move towards working more directly with posterior estimates of the effect size itself.

ADD REPLYlink written 13 days ago by Michael Love1.5k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 1012 users visited in the last hour