Find genes differentially expressed between two conditions

Question

[Lorenzo Fabbri]

Suppose I have a set of genes of interest. These genes, together with their values, come from the CMap project. I'm thus using Z-scores rather than expression profiles. Suppose now I have two matrices: the rows are drugs and the columns are genes. The difference between the two is that the drugs from the first matrix belong to class "A" while the drugs from the second matrix belong to class "B". Is there a method (i.e., a tool) to compare the genes of these two matrices and say whether there is a difference between the two classes? The difficult part is that the CMap provides Z-scores.

[shoujun.gu]

If you think each drug as a single cell, then any single cell DE tools could be used to analyze the difference. But most of the tools need raw counts as input, you may need to read their documents to see whether there is some way to bypass this requirement.

Answer 1

[Kevin Blighe]

Hey,

Having the data on the Z-scale may actually be favourable due to the fact that Z-scores relate to 'standardised' scores, and should follow a normal distribution centered around 0. One assumption that you'd be making is that CMAP has pre-processed the data to remove other sources of bias. Thus, you can use any 'traditional' test, such as [parametric (non-parametric)]:

• Unpaired t-test (Mann-Whitney U test)
• Paired t-test (Wilcoxon Rank Sum test)
• ANOVA (Kruskal-Wallis test)

You could probably fit a generalised linear model to the data, too, of the form:

glm(class ~ gene, family = binomial(link = 'logit'))

..where class is A|B (binary categorical), and gene is a continuous variable comprising the gene expression values.

Kevin