Hypothesis testing and multiple testing - Answer key

Exercise 1

You are studying brain maturation and growth patterns in mouse cortex and have obtained RNA-seq data for a total of 24 mice. These samples were acquired at 2 developmental stages (3 dpf and 10 dpf) and with or without treatment using a growth inhibitor (Monoamine oxidase (MAO) inhibitors). For each developmental stage and treatment combination you have 6 replicates. You also have sex information for these mice (12 males and 12 females).

What steps are necessary to take to decide what your model should be?

Look at QC analyses such as PCA to determine which metadata factors are associated with large amounts of variation in the data.

What is an appropriate hypothesis test if you are testing for expression differences across the developmental stages?

We only have 2 developmental stages, so we can use the Wald test for comparison.

Provide the line of code used to create the dds object.

Here we assume that the metadata includes the following columns: developmental_stage, treatment, and sex. We want to include treatment and sex as covariates in our model, but make sure that developmental_stage is the last factor in the model so that the default result that is returned when using the results() function will be for our main effect.

# create the dds object
dds <- DESeqDataSetFromTximport(txi, colData = meta, design = ~ treatment + sex + developmental_stage)

Provide the line of code used to run DESeq2.

# run DESeq
dds <- DESeq(dds)

Would you use a different hypothesis test if you had 3 developmental timepoints?

In that case we would use a likelihood ratio test, because there are more than two groups for comparison. Since the full model is ~ treatment + sex + developmental_stage, the reduced model is then ~ treatment + sex. (We don’t use ~1 here, because our reduced model does still include factors.)

# create the dds object
dds <- DESeqDataSetFromTximport(txi, colData = meta, design = ~ treatment + sex + developmental_stage)

# run DESeq LRT
dds_lrt <- DESeq(dds, test = "LRT", reduced = ~ sex)