Robust rank aggregation with imprecise p values

Question

[english.server]

Hi!

In robust rank aggregation method of meta-analysis of (low-throughput) gene expression studies, how shall we treat the results with imprecise p values such as (p<0.001, p<0.05)?

I'd also love to know the correct "English" word used for these p values (instead of the word "imprecise" that I used)

Thanks

[Jean-Karim Heriche]

What do you mean by "treat the results"?

[english.server]

I meant assign ranks to them.

gene1 p=0.001
gene2 p<= 0.001
gene3 p = 0.00003

How to rank these gene?

[Jean-Karim Heriche]

You can't rank them meaningfully because you've lost information, e.g. p<0.01 could mean p = 0.000001 or p = 0.0099. However, when people write p < x, this often means p = x - eps where eps is close to 0 relative to x, i.e. p<0.01 means p = 0.0099. Assuming this, you could consider all p < x as p=x-e with e small enough then deal with the ties e.g. by ranking them at random. However, ranking by p-values may not be the best approach because it ignores the magnitude of the effect. Depending on what your goal is, you may consider ranking by some score/measurement instead.

[english.server]

Thank you very much for your response.