Hello Martin,

I think the "correct combination" depends on what you want to find and also is influenced by the data itself. There isn't really a reference workflow.

For example, if you are doing DEA, you could consider the experimental design. It is complex with more than one varying factor? Then you might want to consider methods that use the negative binomial distribution as the basis of the modeling of the RNA-Seq counts (e.g., DESeq, edgeR). I personally like the estimation of within-group variablity of DESeq (something you want to consider when working with less than 5 replicates per group) and experienced that edgeR is more tolerant and therefore reports more genes. On the other hand, non-parametric methods such as SAMSeq or NOISeq do not assume anything about the distribution but rather rank the genes by expression and use statistics to identify the genes that are differential expressed.

But sometimes is also a matter of personal preference. I personally like Segemehl and featureCounts. If you are using DESeq its important to work on the raw data which featureCounts provides. Also if I not mistaken, cufflinks does a FPKM/RPKM normalization and this means it also normalizes in respect to the gene/transcript length which isn't recommended.

Also have a look into the manuals of the tools you might want to use, sometimes the workflow is illustrated with the tools used in up- or downstream analyses. That could help.

Good luck

Rich

Dear Richard A. Schäfer ,

thank you very much for your quick answer, yes, I agree that (maybe unfortunately) the choice of the tools is a matter of personal preference. On the other side, this makes the most of experiments hardly to repeat or reanalyzed, because no one is familiar with all approaches/codes/etc.

RNAseq is growing scientific field, so the unified workflow for basic DE experiments could be a great help to the all wet-lab biologists, which want to analyze their samples on their own :)

Personally I use Galaxy for all my differentional expression RNAseq analyzes and have a great dilemma which combination of tools is the most appropriate option (in the end I am using Trinity -> Kallisto -> EdgeR most often)

Best regards,

Martin

Trinity versions can be useful.

Dear Aditya Banerjee ,

thanks for your answer, Trinity is fine for "assembly" step, but except Trinity there are minimally 4 other programs...Same for "Quantification" step and "Statistical Analyses" step. So there is a minimum of 125 combinations (if we consider 3 steps (assembly, quantification, statistics) and only 5 various tools for each step (which is probably strongly underestimated))

In my modest opinion, there is an urgent need for some reference workflow for basic differential gene expression analyses.

Best regards,

Martin

If you are interested in differential gene expression analysis with some pathway analysis, including quality checks - you can take some ideas from the workflow we use. Some examples can be seen here:

https://github.com/uhkniazi/BRC_FOG1KO_John_PID_20

https://laplacebayes.wordpress.com/category/bioinformatics/

Dear Umar Niazi , thanks for your answer, but I am not familiar with codes and scripts, I use user-friendly Galaxy websites ( https://usegalaxy.eu/). My question was about the most appropriate combination of tools, eg. Tophat - Cufflink - EdgeR or Trinity - Kallisto - DEseq or STAR - featureCounts - BaySeq. In other words, what is currently the best combination of tools for RNAseq data analysis?

Martin, there is no right answer. As you have mentioned, there are many combinations and each of the combinations has its own pros and cons.

You should start by asking yourself what is the main goal of the experiment. Then, you have to setup an appropriate experimental design. Once you have this you can look at the tools you might be using. You have to consider the main question of the experiment, the studied organism, number of replicates, used methodology, and your expertise. There have been many attempts to compare full RNA-Seq pipelines and select "the best" one but it's simply not possible because of the differences between the experiments.

If you want to use something simple for non-bioinformatician you are basically limited to the Galaxy or Chipster (unless you want to pay for the software). In general, running bioinformatics tools doesn't require much coding. If you want to advance a bit you can try Nextflow pipelines (https://nf-co.re/rnaseq) or R pipelines (for example https://www.bioconductor.org/packages/devel/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html).

If you want to get the "best" pipeline you would have to read comparisons/reviews for each step (qc, preprocessing, alignment, postprocessing, counting, differential expression) and select the most suitable tools for your experiment. In case you don't have reference genome/transcriptome (you mentioned assembly) you would have to use Trinity or similar. If this is your case, following with Kallisto and edgeR/DESeq2 sounds ok.

If you want to stick with the public Galaxy you are limited only to the tools available there.

P.S. Forget about TopHat and Cufflinks. These are really outdated. If you want you can use the most up-to-date tools from the same team HISAT2 (replaced TopHat/TopHat2) and StringTie (easier to use than Cufflinks).

Jan Oppelt , thank you very much for your complex answer! I didn´t know about Chipster, it looks great, I will try :-) In my case, there is no reference genome or transcripome...Ok, thanks for the positive feedback, I will continue with Trinity - Kallisto - EdgeR.

Cheers

Martin

We have ready to use pipelines which use the NeatSeq-Flow platform https://neatseq-flow.readthedocs.io/en/latest/ and https://neatseq-flow.readthedocs.io/projects/neatseq-flow-modules/en/latest/ . NeatSeq-Flow can be executed through a friendly GUI or from the command line. Don't hesitate to contact us for further help ([email protected])