ldivide efficiently finds independent LD-blocks from genetic data, using only VCF/BCF files. It is ethnically inclusive in that it relies on no previous data and has no parameters, hidden or otherwise, which would favor already well-studied populations.
It is a reimagined implementation of ldetect.
To make analyses of GWAS results comparable we want LD-blocks to be computed a priori.
A naive and perhaps the most common way to predefine LD-blocks is to split the genome into non-overlapping segments of equal size. This is likely to split regions with strong association signals into multiple blocks.
ldivide works in three steps:
- it slides a window over the SNPs to compute the summed R-squared statistic of each SNP to those within the window. This leads to a vector of association-scores, one per SNP.
- the vector of association scores is smoothed with low-pass filters of increasing width until the desired number of breakpoints are found
- a local search is done within the breakpoints found above to find true local minima
The results from running our pipeline are provided here: [link to come]
- very fast and memory efficient
- easy to use
- nice visualization of the results
- inclusive; does not use previous data or assumptions which would favor already well-studied populations
- few parameters and none hidden;
- the number of SNPs in each window used to compute association signals
- the ones used in the signal processing; which are these?
ldetect (Berisa and Pickrell, 2015) solved this problem by finding blocks where the SNPs had strong association signals to other SNPs within the block and weak to those outside. This method incorporated knowledge about genetic distance when computing association signals. In the supplementaries, we discuss the downsides of this, but the main points are that the genetic recombination data have several problems:
- they are computed on old data
- they are not computed for all populations
- they are not easily reproducible; the data are provided as-is without the code which produced them