ldsep (version 2.1.5)

ldest_comp: Estimates of composite pairwise LD based either on genotype estimates or genotype likelihoods.

Description

This function will estimate the composite LD between two loci, either using genotype estimates or using genotype likelihoods. The resulting measures of LD are generalizations of Burrow's "composite" LD measure.

Usage

ldest_comp(
  ga,
  gb,
  K,
  pen = 1,
  useboot = TRUE,
  nboot = 50,
  se = TRUE,
  model = c("norm", "flex")
)

Value

A vector with some or all of the following elements:

D

The estimate of the LD coefficient.

D_se

The standard error of the estimate of the LD coefficient.

r2

The estimate of the squared Pearson correlation.

r2_se

The standard error of the estimate of the squared Pearson correlation.

r

The estimate of the Pearson correlation.

r_se

The standard error of the estimate of the Pearson correlation.

Dprime

The estimate of the standardized LD coefficient. When type = "comp", this corresponds to the standardization where we fix allele frequencies.

Dprime_se

The standard error of Dprime.

Dprimeg

The estimate of the standardized LD coefficient. This corresponds to the standardization where we fix genotype frequencies.

Dprimeg_se

The standard error of Dprimeg.

z

The Fisher-z transformation of r.

z_se

The standard error of the Fisher-z transformation of r.

p_ab

The estimated haplotype frequency of ab. Only returned if estimating the haplotypic LD.

p_Ab

The estimated haplotype frequency of Ab. Only returned if estimating the haplotypic LD.

p_aB

The estimated haplotype frequency of aB. Only returned if estimating the haplotypic LD.

p_AB

The estimated haplotype frequency of AB. Only returned if estimating the haplotypic LD.

q_ij

The estimated frequency of genotype i at locus 1 and genotype j at locus 2. Only returned if estimating the composite LD.

n

The number of individuals used to estimate pairwise LD.

Arguments

ga

One of two possible inputs:

1. A vector of counts, containing the genotypes for each individual at the first locus. When type = "comp", the vector of genotypes may be continuous (e.g. the posterior mean genotype).

2. A matrix of genotype log-likelihoods at the first locus. The rows index the individuals and the columns index the genotypes. That is ga[i, j] is the genotype likelihood of individual i for genotype j-1.

gb

One of two possible inputs:

1. A vector of counts, containing the genotypes for each individual at the second locus. When type = "comp", the vector of genotypes may be continuous (e.g. the posterior mean genotype).

2. A matrix of genotype log-likelihoods at the second locus. The rows index the individuals and the columns index the genotypes. That is gb[i, j] is the genotype likelihood of individual i for genotype j-1.

K

The ploidy of the species. Assumed to be the same for all individuals.

pen

The penalty to be applied to the likelihood. You can think about this as the prior sample size. Should be greater than 1. Does not apply if model = "norm", type = "comp", and using genotype likelihoods. Also does not apply when type = "comp" and using genotypes.

useboot

Should we use bootstrap standard errors TRUE or not FALSE? Only applicable if using genotype likelihoods and model = "flex"

nboot

The number of bootstrap iterations to use is boot = TRUE. Only applicable if using genotype likelihoods and model = "flex".

se

A logical. Should we calculate standard errors (TRUE) or not (FALSE). Calculating standard errors can be really slow when type = "comp", model = "flex", and when using genotype likelihoods. Otherwise, standard error calculations should be pretty fast.

model

Should we assume the class of joint genotype distributions is from the proportional bivariate normal (model = "norm") or from the general categorical distribution (model = "flex"). Only applicable if using genotype likelihoods.

Author

David Gerard

Examples

set.seed(1)
n <- 100 # sample size
K <- 6 # ploidy

## generate some fake genotypes when LD = 0.
ga <- stats::rbinom(n = n, size = K, prob = 0.5)
gb <- stats::rbinom(n = n, size = K, prob = 0.5)
head(ga)
head(gb)

## generate some fake genotype likelihoods when LD = 0.
gamat <- t(sapply(ga, stats::dnorm, x = 0:K, sd = 1, log = TRUE))
gbmat <- t(sapply(gb, stats::dnorm, x = 0:K, sd = 1, log = TRUE))
head(gamat)
head(gbmat)

## Composite LD with genotypes
ldout1 <- ldest_comp(ga = ga,
                     gb = gb,
                     K = K)
head(ldout1)

## Composite LD with genotype likelihoods
ldout2 <- ldest_comp(ga = gamat,
                     gb = gbmat,
                     K = K,
                     se = FALSE,
                     model = "flex")
head(ldout2)

## Composite LD with genotype likelihoods and proportional bivariate normal
ldout3 <- ldest_comp(ga = gamat,
                     gb = gbmat,
                     K = K,
                     model = "norm")
head(ldout3)