Single-Tissue eQTLs: P-value threshold in GTEX ??

Question

[GabrielMontenegro]

Hi all,

eQTL is not my area of research, however I am interested to see if a particular SNP has a significant expression in the GTEX database.

I have queried this SNP rs4888622 (found on a paper on gene expression on the human brain). This is the result: (link: https://www.gtexportal.org/home/snp/rs4888622)

Gencode Id  Gene Symbol Variant Id  P-Value Effect Size Effect Size
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 8.0897e-20  -1.09348    -1.1
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 7.24774e-17 -1.0366 -1.0
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 1.43283e-16 -1.12875    -1.1
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 1.02893e-13 -0.900694   -0.90
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 3.07218e-12 -1.00953    -1.0
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 9.55899e-11 -0.973495   -0.97
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 3.60757e-8  -0.780061   -0.78
ENSG00000140873.11  ADAMTS18    16_77396849_T_C_b37 2.51812e-7  -0.786018   -0.79

Now, the P-value is highly significant (top is 8.0897e-20) in many of this tissue, but what P-value threshold should I use? Should I use for a Bonferroni correction based on the number of SNPs and tissues there? How do I get these numbers?

Thank you in advance.

[dongguiying2017]

I also have this question, so what the p value should be used when consider the eQTL?

Answer 1

[Kevin Blighe]

You will have to calculate the nominal p-value at which the FDR (false discovery rate) is 5%, and then use that as the cut-off. You can choose other thresholds too, of course. Prior to the association testing, your input SNPs should also be filtered based on HWE (Hardy-Weinberg Equilibrium), e..g, filter out those with HWE p < 5 × 10^−6.

My answer is to merely point you to other material to help link up some of these other posts and work in the field:

• Power, false discovery rate and Winner’s Curse in eQTL studies
• A: Permutation Based Thresholds For Eqtl Analysis?
• MatrixeQTL FDR and Q-Qplot

Kevin