How to understand pval_nominal_threshold?

[Geneva0725]

Hi,

Thanks for developing this awesome tool. I am confused with pval_nominal_threshold column outputted by post.calculate_qvalues function. Is this column a significant threshold? Can I filter significant result with qval < 0.05 or pval_nominal < pval_nominal_threshold?

[francois-a]

The qval column should be used to filter for, e.g., eGenes. The pval_nominal_threshold column indicates for each phenotype what the largest nominal p-value that still passes FDR would be. The calculation is performed here.

[Geneva0725]

Thanks for replying!

But I found that in my results some eQTL's nominal p-value less than nominal p-value threshold, but its qval is greater than the threshold, likeqval < 0.1, and I show an example below. Do you know how that happens? And I'm wondering if I can relax the significance threshold to pval_nominal < pval_nominal_threshold?

phenotype_id	num_var	beta_shape1	beta_shape2	true_df	pval_true_df	variant_id	tss_distance	ma_samples	ma_count	af	pval_nominal	slope	slope_se	pval_perm	pval_beta	qval	pval_nominal_threshold
HNRNPK	4224	1.0438424	326.04254	18.688456	1.6377151131966172e-06	chr9_84582310_C_A	601694	21	28	0.53333336	3.6158298851078376e-07	-1.0103301	0.13879848	0.00019998	0.00037639	0.2648134184906671	9.806876908439229e-07

[francois-a]

There are some edge cases where this will happen (partially because of how this threshold computed using the phenotype closest to the FDR cutoff). pval_nominal_threshold is not meant to be used in any downstream analyses. To identify significant (independent) signals for each phenotype, fine-mapping should be used instead.