Allow per phenotype p-value thresholds

[edg1983]

The idea

Currently, the p-value thresholds for SNPs selection and locus breaker are configured per study directly in the summary stat input table (p_thresh1 and p_thresh2 columns). Thus, the same thresholds are applied to all phenotypes for a given study.

However, this is not ideal for molecular traits (like eQTLs or pQTLs results) where one may prefer to set different thresholds for each gene / protein.

A possible implementation

• We allow an additional optional input column in the summary stat input table, like p_thresh_table. This must point to a TSV table with columns: study_id, pheno_id (aka gene_id), p_thresh1, p_thresh2
• We update the locus-breaker R script to accept an extra argument --p_thresh_table.
• When we load the summary stats, we add 2 columns representing the thresholds. These are populated with fixed values from --p_thres1 and --p_thres2 then --p_thresh_table is not set, otherwise we use a merge by (study_id, pheno_id) to populate them with information from the p threshold table
• SNPs are filtered based on the p_thresh columns (and then these can be removed eventually)

We have to point out in the docs that when the p-value threshold table is provided, this will take precedence over the fixed values.

[shoffm]

Thanks so much for adding this issue! We've been thinking more about this and have an idea of an alternative solution. Instead of having to update the whole pipeline to accommodate per-gene p-value thresholds for molQTL studies, what if we just include each gene as a different “study”/row in the input table?

One limitation we can foresee is that it would then re-write the genotyping file ~20k times across 20k genes, because when we ran flanders across cell types from the same dataset, it re-made and over-wrote the same genotyping file for each cell type even though they were all the same. Would it therefore be easier to write a check for whether the genotyping file exists?

Are there any other limitations you can see including each gene in a molQTL as separate studies?