Question

How can I know the appropriate p-value threshould?

1

Entering edit mode

8.1 years ago

marwa.cs91 ▴ 30

Hello all,

I used a dataset of patients and normal individuals. This dataset contains 1560242 SNPs after quality control. I used the genome-wide significance (p < 5e -8). The number of SNPs after using the threshold (p < 5e -8) are very small. How can I use another proper threshold to get more SNPs ?

thanks

p-value • 3.1k views

ADD COMMENT • link 8.1 years ago by marwa.cs91 ▴ 30

1

Entering edit mode

Hint: correct for multiple testing.

ADD REPLY • link 8.1 years ago by Devon Ryan 104k

0

Entering edit mode

Thanks for your important hint. I used plink --adjust command to know FDR significance values. How can I know the significance threshold of FDR? thanks

ADD REPLY • link 8.1 years ago by marwa.cs91 ▴ 30

1

Entering edit mode

It's the same arbitrary threshold as with p-values.

ADD REPLY • link 8.1 years ago by Devon Ryan 104k

0

Entering edit mode

can I use the threshold FDR <0.05 Is it true ?

thanks

ADD REPLY • link 8.1 years ago by marwa.cs91 ▴ 30

2

Entering edit mode

The FDR represents the expected number of false positives in your data so if you're happy with 5% false positives then 0.05 is a good threshold. It depends on what you want to do.

ADD REPLY • link 8.1 years ago by Jean-Karim Heriche 27k

0

Entering edit mode

Thanks Jean-Karim for your description . My dataset after quality control contains 1560242 SNPs but after using genome-wide significance(p < 5e -8) , the number of SNPs became very small . I tried bonferroni correction (0.05/total number of snps) but these threshold gave also very small number of SNPs.

Could you tell me suggestion.

Thanks and I really appreciate any help you can provide

ADD REPLY • link 8.1 years ago by marwa.cs91 ▴ 30

1

Entering edit mode

The Bonferroni correction is generally considered too conservative for genomics studies which is why the FDR is usually preferred. How you choose the threshold on what you consider significant depends on what you want to do with the SNPs afterwards. Another way of looking at it is to consider the trade off between false positives and false negatives. If you don't care about false negatives but don't want false positives, only keep very low FDR-corrected p-values but if on the other hand, you want to minimize the risk of missing relevant SNPs, you have to accept more false positives by using a higher cut-off. Having few significant SNPs is not necessarily bad e.g. when you need to pursue each one with further experiments.