Output Options for ldfast()

[dcgerard]

It would be great if

1. ldfast() kept SNP names (as row and column names)
2. ldfast() allowed for a return type of data.frame
3. ldfast() had an option to keep SNP locations.

Thanks to Heather Tuttle for the suggestion.

As a workaround until I make this a feature, you can use:

library(updog)
library(ldsep)
library(reshape2)

## Get data and fit ldfast ----
data("uit", package = "ldsep")
gp <- format_multidog(x = uit, varname = paste0("Pr_", 0:4))
ldout <- ldfast(gp = gp, type = "r2")

## Format to data frame ----
colnames(ldout$ldmat) <- rownames(ldout$ldmat) <- dimnames(gp)$snp
ldout$ldmat[lower.tri(ldout$ldmat, diag = TRUE)] <- NA
lddf <- melt(ldout$ldmat, na.rm = TRUE)
lddf

You can then use a tool like merge() or dplyr::left_join() to add genomic positions to this data frame.

[dcgerard]

To finish off this example, and provide code for an LD decay plot (as also suggested by @lfelipe-ferrao), here is some quick and dirty code:

library(dplyr)
library(ggplot2)
library(scam) ## monotone constraint

## Suppose we have the following genotype locations
dfloc <- data.frame(SNP = rownames(ldout$ldmat), loc = 1:10)

## Merge LD estimates
lddf |>
  left_join(dfloc, by = c("Var1" = "SNP")) |>
  left_join(dfloc, by = c("Var2" = "SNP")) |>
  rename(SNP1 = Var1, SNP2 = Var2, r2 = value, Loc1 = loc.x, Loc2 = loc.y) |>
  mutate(dist = abs(Loc1 - Loc2)) |>
  distinct(SNP1, SNP2, .keep_all = TRUE) ->
  lddf

## Make plot
ggplot(data = lddf, mapping = aes(x = dist, y = r2)) +
  geom_point() +
  geom_smooth(method = scam, formula = y ~ s(x, bs = c("mdcx", k = 8)), se = FALSE)