select_IV_bs: Select instrumental variables for the MR-MtRobin analysis...

In kjgleason/Mr.MtRobin: A Multi-Tissue transcriptome-wide Mendelian Randomization method ROBust to INvalid instrumental variables.

Description

Selects a set of instrumental variables (IVs) based on specified criteria to be used in two-sample Mendelian Randomization analysis using MR-MtRobin. After restricting the candidate SNPs to those with P-value less than pval_thresh in at least nTiss_thresh tissues, the function iteratively removes one SNP/variant from the pair with the largest pairwise LD r^2 until all pairwise LD r^2 among remaining SNPs is less than ld_thresh. The variant selected to be removed is based on the ranked criteria 1) greatest number of pairwise LD r^2 > ld_thresh with other remaining SNPs; 2) fewest number of tissues with P-value less than pval_thresh; and 3) largest minimum P-value of association with expression of geneID.

Usage

select_IV_bs(geneID, eqtl_data, nTiss, LD, ld_thresh = 0.5,
  pval_thresh = 0.001, nTiss_thresh = 3)

Arguments

geneID	character string of the gene to be tested.
eqtl_data	data.frame of summary statistics from eQTL study.
nTiss	integer of the number of tissues analyzed by the eQTL study.
LD	matrix of LD correlation coefficients (r, not r^2).
ld_thresh	numeric of pairwise LD threshold (r^2) .
pval_thresh	numeric of P-value threshold for SNP-tissue pair to be used as IV.
nTiss_thresh	integer of minimum number of tissues in which a candidate IV must have P-value less than pval_thresh.

Details

The following are additional details describing the input arguments.

The data.frame eqtl_data should have the character variables gene_id and variant_id (as identifiers), as well as the variables pvalue_j and beta_j for j in {1,...,nTiss}, corresponding to the P-value testing the null hypothesis of no association between gene_id and variant_id and the estimated coefficient (beta) of the marginal effect of variant_id on gene_id expression in each respective eQTL analysis, respectively. Note that the names of the p-value and beta columns must match the conventions pvalue_j and beta_j for j in {1,...,nTiss} exactly.

Note that the matrix LD should hold correlation coefficients (i.e. r), not their squared values (r^2).

Value

A character vector of the SNP/variant identifiers to be used as instrumental variables with geneID.

Examples

select_IV_bs(geneID="gene1", eqtl_data=eqtl_stats_example, nTiss=10, LD=LD_example)

kjgleason/Mr.MtRobin documentation built on Sept. 6, 2021, 5:13 p.m.