Why we use adjusted p-value not p-value in differential gene expression analysis ?

Question

[Amr]

Why we use adjusted p-value not p-value in differential gene expression analysis?

Answer 1

[Asaf]

Because we need to correct for multiple hypotheses. If you ask a lot of questions, about 5% of them will pass the 0.05 p-value threshold just by chance.

[Istvan Albert]

One important detail is also that some adjusted p-values are not even p-values in the original sense.

For example, an FDR may also be called an adjusted p-value, but it is a different concept altogether.

[Amr]

@Istvan Thanks for your useful comment

[Amr]

Do you want to say that the adjusted p-value margin is bigger so it allows more genes to be under the hypothesis of 0.05? I mean for example, 100 genes are statistically significant less than 0.05 but for adjusted p-value calculations with Benjamini- Hochberg there will be more than 100 genes statistically significant with the same threshold 0.05? If you don't mean what I mentioned or if I am wrong, then please correct me...Thanks

[liorglic]

Adjusted p-values will always be equal or larger than original p-values, so the situation is opposite to what you wrote: there will probably be less than 100 significant genes left.

[Istvan Albert]

Adjusted p-values will always be equal or larger than original p-values

that is not quite right - an FDR type of adjustment could make some p-values smaller - though I do also argue in this thread that calling an FDR a p-value is a bit of a misnomer.

But I want to make another point. If we wanted to be rigorous, there is no such thing as an adjusted p-value in the first place - what we adjust is the significance level,

For example, if the significance level is 0.05 before adjusting, that level becomes 0.005 after adjustment. So the correction should be applied to the significance levels, not the p-values.

But to make things "more clear", in typical data analysis, we end up multiplying the p-value with a factor and still using 0.05 as the significance - but to my understanding, that can be misleading when we start to interpret p-values in different contexts.

[Amr]

@liorglic & @Asaf Thanks a lot for clarification.