Normalization module

This module allow to normalize expression module results.

  • Name: normalization
  • Available: Only in local mode

  • Input port:
    • input: expression file in TSV format (format: expression_results_tsv)

  • Output: Control graphics, LaTeX report and count matrices (raw, pooled and normalized)

  • Optional parameters
  • Parameter Type Description Default value
    r.execution.mode string The R execution mode. The available mode values are: process, rserve and docker. process
    rserve.servername string The Rserve server name to use in rserve execution mode not set
    docker.image string The Docker image to use in Docker execution mode. genomicpariscentre/deseq:1.8.3
  • Configuration example:
  • <!-- Normalization step -->
    <step id="mynormlizationstep" skip="false" discardoutput="false">
    	<name>normalization</name>
    	<parameters />
    </step>
    

Required R packages installation

Eoulsan normalization module use R with DESeq and FactoMineR packages as statistical backend. Normalization module was tested with R 2.15, DESeq 1.8.3 and FactoMineR_1.20. You need to install two R packages on your computer or Rserve server:

$ sudo R
> source("http://bioconductor.org/biocLite.R")
> biocLite("DESeq")
> install.packages("FactoMineR")
			

Interpreting output files

Null count proportion barplot

Null count proportion barplot
This barplot represent the null count proportion in all sample. This gives a first idea of expression differences.

Total count barplot

This graph represent the total counts for each sample. It also give a first idea of expression differences.

Raw data (before technical replicates pooling and normalization)

Raw total count barplot

After technical replicates pooling

Pooled total count barplot

log2(counts + 1) distribution boxplot

This type of graph is useful to see if normalization was performed well. When all boxes are aligned or at least the median, the normalization worked well.

On raw data

Raw boxplot

After technical replicates pooling

Pooled boxplot

After normalization

Normalized boxplot

log2(counts + 1) distribution density

This graph show the distribution profiles of each sample. It is useful to verify that technical replicates count distributions are close enough to pool them and to see if normalization have corrected well distribution differences.

On raw data

Raw density plot

After technical replicate and normalization

Normalized density plot

Clustering dendrogram

This graph is plotted with hclust R function with the ward method and the distance used is 1-(correlation/2). It it useful to see if replicates are grouped together.

On raw data

Raw cluster dendrogram

After technical replicate and normalization

Normalized cluster dendrogram

PCA scatter plot

This graph have the same goal than the cluster dendrogram but it is more easy do read.

On raw data

Raw PCA plot
Warning : as we can see on this graph if some of the samples have a number of total count very higher than the others. The first dimension of the PCA is the count number and the graph isn't really informative in this case.

After technical replicate and normalization

Normalized PCA plot
On this graph like on the corresponding dendrogram we can see that samples of MO_injection and WT_injection are grouped together which is not waiting. Actually, in this experiment injection was performed at 2 different times and these two conditions are control conditions of 2 strains, so these graph shows that there is an "injection time effect" stronger than the difference between the 2 strains.

Count matrix

This module also provide a raw count matrix and a normalized count matrix which can be use for other analysis.