Genomatix RNA-Seq workflow

[rebrendi]

Hello,

has anyone used Genomatix for RNA-Seq analysis?
I am particularly interested in the interpretation of the output of the Audic-Claverie algorithm for differential gene expression analysis described here: http://www.genomatix.de/online_help/.../claverie.html
Can someone tell me how is calculated the "NE value" for differential expression?
How it relates to RPKM?

Thank you!

[Bernward]

Hi,

The NE is explained at:

http://www.genomatix.de/online_help/help_regionminer/RNA_clustering.html

Normalized expression/enrichment value (NE-value)
The NE-value is calculated based on the following formula:

NE = c * #readsregion / (#readsmapped * lengthregion)

where NE is the normalized expression or enrichment value,
#readsregion: the reads (sum of base pairs) falling into either the transcript or the cluster region,
#readsmapped: all mapped reads (in base pairs),
lengthregion: the transcript or cluster length in base pairs
and c a normalization constant set to 10E7.

This is quite similar to RPKM.

Regarding Audic & Claverie statistics: This was originally designed for the count data in cDNA / Sage libraries. "The Significance of Digital Gene Expression Profiles" Audic & Claverie, 1997.
The idea is to calculate the conditional probability to find y counts for your sequence feature (gene, transcript, ChIP region) in your treatment data when you saw x counts in the control for the same sequence feature.

However, for NGS data this calculation overestimates the number of significantly deviant features, since it is based on a Poisson model. As discussed here several times the Poisson model doesn't fit well for NGS data.

If replicates are available I strongly recommend using DESeq or EdgeR. For ChIP-seq clusters however, you must merge or intersect the clusters from your replicates. The workflow is explained at:

http://www.genomatix.de/online_help/help_regionminer/ChIPSeqWorkflow.html

Regards,
Bernward