Differential Expression for RNA-Seq

The Differential Expression for RNA-Seq tool performs a statistical differential expression test for a set of Expression Tracks. It uses multi-factorial statistics based on a negative binomial GLM. The tool supports paired designs and can control for batch effects. The statistical analysis is described in more detail in The statistical model.

To run the Differential Expression for RNA-Seq analysis:

        Toolbox | RNA-Seq and Small RNA Analysis (Image rna_seq_group_closed_16_n_p)| Differential Expression (Image rna_expression_folder_closed_16_n_p) | Differential Expression for RNA-Seq (Image dge_rnaseq_16_n_p)

Select a number of Expression tracks (Image rnaseqtrack_16_h_p) and click Next figure 31.76.

Image expressionrnaseq
Figure 31.79: Select a number of Expression tracks.

For Expression Tracks (TE), the values used as input are "Total transcript reads". For Gene Expression Tracks (GE), the values used depend on whether an eukaryotic or prokaryotic organism is analyzed, i.e., if the option "Genome annotated with Genes and transcripts" or "Genome annotated with Genes only" is used. For Eukaryotes the values are "Total Exon Reads", whereas for Prokaryotes the values are "Total Gene Reads".

Note that the order of comparisons can be controlled by changing the order of Expression track inputs.

The available normalization options can be seen in figure 31.77.

Image normamethod
Figure 31.80: Normalization methods.

First, choose the application that was used to generate the expression tracks: Whole transcriptome RNA-Seq, Targeted RNA-Seq, or Small RNA. For Targeted RNA-Seq and Small RNA, you can choose between two normalization methods: TMM and Housekeeping genes, while Whole transcriptome RNA-Seq will be normalized by default using the TMM method. For more detail on the methods see TMM Normalization.

TMM Normalization (Trimmed Mean of M values) calculates effective libraries sizes, which are then used as part of the per-sample normalization. TMM normalization adjusts library sizes based on the assumption that most genes are not differentially expressed.

Normalization with Housekeeping genes can be done when a set of housekeeping genes to use is available: in the "Custom housekeeping genes" field, type the name of the genes separated by a space. Finally choose between these two options:

When working with Targeted RNA Panels, we recommend that normalization is done using the Housekeeping genes method rather than TMM. Predefined list of housekeeping genes are available for samples generated using Human and Mouse QIAseq panels (hover with the mouse on the dialog to find the list of genes included in the set). If you are working with a custom panel, you can also provide the corresponding set of housekeeping genes in the "Custom housekeeping genes" as described above.

In the Experimental design panel (figure 31.78), a Metadata table must be selected that describes the factors and groups for all the samples.

Image experimental_design
Figure 31.81: Setting up the experimental design and comparisons.

The Comparisons panel determines the number and type of statistical comparison tracks output by the tool (see Output of the Differential Expression for RNA-Seq tool for more details).

The Differential Expression for RNA-Seq tool produces different numbers and types of statistical comparison tracks depending on the settings of the Comparisons panel. Depending on the choice either a Wald test or a Likelihood Ratio test is used. For example, assume that we test a factor called 'Tissue' with three groups: skin, liver, brain.

Note: Fold changes are calculated from the GLM, which corrects for differences in library size between the samples and the effects of confounding factors. It is therefore not possible to derive these fold changes from the original counts by simple algebraic calculations.

In the final dialog, choose whether to downweight outlier expressions, and whether to filter on average expression prior to FDR correction.

Downweighting outliers is appropriate when a standard differential expression analysis is enriched for genes that are highly expressed in just one sample. These genes do not fit the null hypothesis of no change in expression across samples. Downweighting comes at a cost to precision and so is not recommended generally. For more details, see Downweighting outliers.

Filtering maximizes the number of results that are significant at a target FDR threshold, but at the cost of potentially removing significant results with low average expression. For more details, see Filtering on average expression.