Very low p-values

Question

[ab123]

Hey guys,

Just wondering if I should worry getting some results running the lmfit in limma which reach down to e-17 region for P-Values. That's not the case for all datasets, but it is for two of them.

Is this normal? I know that we can expect some low P-values in expression array datasets, but e-17? Really?

Just to clarify some of the pre-steps include bgCorrection via Limma and then between array normalization.

Thank you!

Answer 1

[Jean-Karim Heriche]

Such a low p-value means that if the null hypothesis were true, it is extremely unlikely to see this value. You don't say what this p-value is associated with but I guess it is change in expression level. This would mean that the change observed is extremely unlikely to be observed if the null hypothesis of no change is true. Usually when I get strong outliers in an experiment, I start worrying about artefacts before considering the effect is real. I would look at the effect size i.e. in your case the magnitude of the change and see if there is an obvious explanation for it.

[ab123]

Thank you Jean! Do you have a suggestion for an approach in limma for checking this?

[Jean-Karim Heriche]

What I am suggesting is not to play more with the computer but to think about your data. Just check the expression value associated with the very low p-values and what samples/genes they are associated with. Ask yourself: Is there a simple explanation for this value for this gene in this sample ? If you have replicates, is it reproducible ? Could it be explained by some technical step or some obvious biology (e.g. it's a cell death marker and you have plenty of dead cells) ?

[ab123]

Many thanks. That makes sense.

[WouterDeCoster]

If this answer was helpful you should upvote it, if the answer resolved your question you should mark it as accepted.