For updates on the entire workflow published in Plant Germline Development (Methods in Molecular Biology), visit:
RNA-Seq data analysis protocol: combining in-house and publicly available data
I wrote a collection of wrapper functions which will be used in the workflow as well. Independent of the workflow, I added several generic examples for RNAseq data analysis.
If you are on an Ubuntu-like system and if you would like to have the newest version of R:
### install the newest version of R
## add the repository to your software sources
# sudo add-apt-repository "deb http://<MIRR>/bin/linux/ubuntu <VERS>/"
# <VERS>: Ubuntu version (code name; e.g. "trusty" for Ubuntu 14.04)
# <MIRR>: A mirror listed on https://cran.r-project.org/mirrors.html
sudo add-apt-repository "deb http://stat.ethz.ch/CRAN/bin/linux/ubuntu trusty/"
## add the authentication key
sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys E084DAB9
## update the software package list
sudo apt-get update
## install R
sudo apt-get install r-base r-base-dev r-cran-rjavaRNAseqWrapper imports several other packages. Install the following dependencies within R:
source("http://bioconductor.org/biocLite.R")
biocLite()
biocLite(c("biomaRt", "DESeq2", "edgeR", "limma", "XLConnect", "gplots", "colorRamps", "SRAdb"))Download RNAseqWrapper and install it:
install.packages("/path/to/RNAseqWrapper_0.99.1.tar.gz", repos=NULL)
# you may need to specify the type:
install.packages("/path/to/RNAseqWrapper_0.99.1.tar.gz", repos=NULL, type="source")Please note that the original version of the package (used in the workflow) is available here: RNAseqWrapper 0.99.0. The newer version contains a bugfix which was caused by an update in DESeq2. At least with DESeq2 >= 1.16.1, you should use the new version of RNAseqWrapper.
Please note that the examples are based on real data, which however is not yet publicly available. Thus the examples are tested, but I can't supply the test data. Please contact me if you encounter a problem.
In the most simple case you may have only two groups of samples (e.g. wild-type vs mutant, treated vs mock control or tissue A vs tissue B):
However, you might have processed the samples on different days (hopefully on each day from both conditions...). In this cases it's sometimes beneficial to include this information as a batch effect:
two group comparison with batch effect
or a "real-world example": paired data from the cancer genome atlas
Eventually you have more than two groups for which you would like to do all pairwise comparisons:
multiple two group comparisons [REQUIRES EACH GROUP TO BE REPLICATED]
However, you might have processed these samples as well in batches:
multiple two group comparisons with batch effect [REQUIRES EACH GROUP TO BE REPLICATED]
You may have two factors in your experiment, e.g. genetic background (wildType vs mutant) and drug treatment (mock vs drug). For cases where one (or both) of the factors have more than two levels (e.g. many different mutants), I recommend to use the approach below. However, in case of two two-level-factors we may also use a crossed factorial design (with/without batch):
two-by-two factorial design with batch effect
Finally, your experiment may have several factors with two or more levels. Multifactorial models tend to be quickly quite complex and non-intuitive. Alternatively one can combine all factors and their levels into one single factor with multiple levels. The comparisons of interest can then be done using linear contrasts:
single/compound factor with several levels [SOME GROUPS MAY BE UNREPLICATED]
(there is no simple one-fits-all solution for the single/compound factor with several levels plus batch effect - well, one can write it out, but it's really lengthy then. And finally - using a model with batch effect is not the best choice in all cases. It's mainly beneficial if the batch effect is strong. Otherwise it is safe (and sometimes better) not to use the batch factor)
- I don't have any replicates, what can I do now?
- edgeR and limma will not work, use the individual DESeq2 functions (the single/compound factor should work in any case as long as there are some groups with replicates). But - just because DESeq2 may run, it does not mean that you should do it.
- think about a possible pseudoreplication. It's not very clean, but you sometimes you have the option to treat two different samples as replicates.
- if it was your doing, don't do it again ;)
- My count table does not have counts per genes but counts per transcripts - is there anything I need to consider?
- Yes - in the function
f.write.DEGtabs.to.workbook()you should setaddGeneCol=TRUE.
- Yes - in the function
The results of the DE analysis are stored as c.DEGtab object. This object is implemented as a reference class and has the following functions associated with it:
DETAB # a c.DEGtab object
# add more data to the tables (this would be saved in the Excel workbooks as well):
DETAB$add_data(x) # x must be a data.frame - merging is done based on rownames()
# convert the DETAB to a regular data.frame:
deData <- DETAB$get_table()
# similar, but round, sort and apply some formatting:
deData <- DETAB$get_print_table()
# extract a list with three gene sets (upregulated in any, conditionA or conditionB):
sigGenes <- DETAB$get_significant_entries(rawPcut = 1, # raw P-value cutoff
adjPcut = 0.05, # FDR cutoff
LFCcut = 0) # LFC threshold
# get some simple information on the number of genes upregulated in one or the other
# condition. The function returns a 3D array with the third dimension being conditionA,
# conditionB and "simple".
stats <- DETAB$get_stats()
# or print them directly:
DETAB$print_stats()
# (markdown is possible as well)
DETAB$print_stats_markdown()