Satistical test : Determine specific association of pathway with a particular stress condition

Question

[khhgng]

Hello everyone,

I have a table in following format- different stress conditions with time points and pathway data :

condition Stress1 Stress2 Stress3

Pathway1

Pathway2

Pathway3

Pathway4

I have the data of no. of genes of each pathway under each condition.

What could be the statistical ways to find if any pathway is specifically associated with a particular stress condition in this data ?

Thanks.

Answer 1

[Jean-Karim Heriche]

Do a chi-squared test (or Fisher's exact test if the counts are small) of independence on the multiway table to test whether pathways are independent from stress then if you consider the test significant, do post-hoc tests on the relevant pairwise comparisons. Don't forget to correct for multiple testing. For more details, see the handbook of biological statistics.

[khhgng]

Thanks. However, since the no. of annotated genes vary between pathways, the counts range from small to large. Should Fisher's exact test be a better choice for this ?

And secondly, when the multiple testing correction should be applied, as a post-hoc test or before ?

[Jean-Karim Heriche]

How small is small is hard to tell without actual numbers but you can't go wrong doing exact tests, it may just take longer to compute. Multiple testing correction is done for the post-hoc tests since that's where you're doing more than one test.