Differential Expression (PubWeb)

Workflow for performing differential expression analysis from gene abundance data

Input Data

The input data for this workflow must consist of two files, a counts table and a manifest table. The counts table contains the number of reads per gene across all specimens, while the manifest contains annotations on those specimens which are used to identify which groups of specimens will be compared.

In addition to the input files, the user must also specify a column of the manifest which will be used to identify which specimens should be compared.

Counts Table Format

The counts table must be formatted as either a TSV or CSV, with the file extension either '.tsv[.gz]' or '.csv[.gz]', as appropriate.

The first column in the counts table must contain the gene ID. Every specimen defined in the manifest must correspond to a column in the counts table. There may be additional columns in the counts table, but they will be ignored in this analysis.

Manifest Table Format

The manifest table must be formatted as a CSV, with the first column containing the specimen names. Every additional column defines the annotations used for those specimens. It is perfectly acceptable to leave a value empty for a particular piece of metadata, in which case the specimen will be ignored in any analysis which uses that feature for comparisons.

Defining Comparisons

Comparisons can be made between specimens using either continuous or categorical variables. A column in the manifest is treated as a continuous variable when all of the values are numeric, and it is treated as a categorical variable when those values are strings.

The user can define the exact comparisons which are tested using a combination of the parameters:

• comp_col: Column used for comparison
• comp_ref: For categorical comparisons, the label used for the reference group
• group_cols: Comma separated list indicating any columns which contain batch information which will be normalized before testing the comparison group
• filter: Optional expression used to filter specimens used for the comparison

Note: Interaction terms are not currently supported.

Consider the following example:

---

specimen	batch	genotype	age
A1	A	wt	16
A2	A	wt	11
A3	A	nod2	14
A4	A	nod2	18
B1	B	wt	10
B2	B	wt	8
B3	B	nod2	6
B4	B	nod2	12

---

Example 1:

To simply compare the wt vs. nod2 samples (while indicating that wt is the reference group), the user would specify:

• comp_col: genotype
• comp_ref: wt

In the results, the fold change will indicate the average value of nod2 / wt for all genes, while using a comp_ref: nod2 would yield the inverse ratio wt / nod2.

Note: Any time a categorical comparison is used, the reference value must be specified to ensure complete clarity.

Example 2:

To test for changes in gene expression as a function of age, the user would specify:

• comp_col: age

Note that comp_ref is not needed for continuous (numeric) variables. In this case, the results will indicate the average change in gene expression for every unit change in age across the dataset.

Example 3:

To test for changes in gene expression as a function of age while controlling for any differences correlated with genotype, the user would specify:

• comp_col: age
• group_cols: genotype

Example 4:

To test for changes in gene expression as a function of genotype while controlling for any differences correlated with age, the user would specify:

• comp_col: genotype
• comp_ref: wt
• group_cols: age

Example 5:

In some cases, the user may want to control for multiple factors, such as the batch indicated in the example above. This can be specified by the user by listing those columns as a comma-separated list.

For example, to test for changes in gene expression as a function of genotype while controlling for any differences correlated with age, while also controlling for batch effects, the user would specify:

• comp_col: genotype
• comp_ref: wt
• group_cols: age,group

Filtering Example 1:

To compare the wt vs. nod2 samples using only the specimens from batch A, the user would specify:

• comp_col: genotype
• comp_ref: wt
• filter: batch == 'A'

Note the use of single quotes used for string matching in filter.

Filtering Example 2:

To compare the wt vs. nod2 samples using only the specimens where age is greater than or equal to 10, the user would specify:

• comp_col: genotype
• comp_ref: wt
• filter: age >= 10

Any boolean expression can be used in the filter parameter, but set membership (e.g. age in [10, 12, 17] is not supported).