New coloc function, builds on coloc.abf() by allowing for multiple independent causal variants per trait through conditioning or masking.
coloc.signals(
dataset1,
dataset2,
MAF = NULL,
LD = NULL,
method = c("single", "cond", "mask"),
mode = c("iterative", "allbutone"),
p1 = 1e-04,
p2 = 1e-04,
p12 = NULL,
maxhits = 3,
r2thr = 0.01,
pthr = 1e-06
)data.table of coloc results, one row per pair of lead snps detected in each dataset
a list with specifically named elements defining the dataset
to be analysed. See check_dataset for details.
as above, for dataset 2
Common minor allele frequency vector to be used for both dataset1 and dataset2, a shorthand for supplying the same vector as parts of both datasets
required if method="cond". matrix of genotype correlation (ie r, not r^2) between SNPs. If dataset1 and dataset2 may have different LD, you can instead add LD=LD1 to the list of dataset1 and a different LD matrix for dataset2
default "" means do no conditioning, should return similar to coloc.abf. if method="cond", then use conditioning to coloc multiple signals. if method="mask", use masking to coloc multiple signals. if different datasets need different methods (eg LD is only available for one of them) you can set method on a per-dataset basis by adding method="..." to the list for that dataset.
"iterative" or "allbutone". Easiest understood with an example. Suppose there are 3 signal SNPs detected for trait 1, A, B, C and only one for trait 2, D.
Under "iterative" mode, 3 coloc will be performed:
* trait 1 - trait 2
* trait 1 conditioned on A - trait 2
* trait 1 conditioned on A+B - trait 2Under "allbutone" mode, they would be * trait 1 conditioned on B+C - trait 2 * trait 1 conditioned on A+C - trait 2 * trait 1 conditioned on A+B - trait 2
Only iterative mode is supported for method="mask".
The allbutone mode is optimal if the signals are known with certainty (which they never are), because it allows each signal to be tested without influence of the others. When there is uncertainty, it may make sense to use iterative mode, because the strongest signals aren't affected by conditioning incorrectly on weaker secondary and less certain signals.
prior probability a SNP is associated with trait 1, default 1e-4
prior probability a SNP is associated with trait 2, default 1e-4
prior probability a SNP is associated with both traits, default 1e-5
maximum number of levels to condition/mask
if masking, the threshold on r2 should be used to call two signals independent. our experience is that this needs to be set low to avoid double calling the same strong signal.
if masking or conditioning, what p value threshold to call a secondary hit "significant"
Chris Wallace