calc_PVEnei {rNeighborGWAS}

Calculating phenotypic variation explained by neighbor effects

Description

A function to calculate PVE by neighbor effects for a series of neighbor distance using a mixed model.

Usage

calc_PVEnei(
  pheno,
  geno,
  smap,
  scale_seq,
  addcovar = NULL,
  grouping = rep(1, nrow(smap)),
  response = c("quantitative", "binary"),
  n_core = 1L
)

Arguments

pheno

A numeric vector including phenotypes for individuals

geno

An individual x marker matrix. Bialleles (i.e., A or a) must be converted into -1 or 1 digit.

smap

A matrix showing a spatial map for individuals. The first and second column include spatial points along an x-axis and y-axis, respectively.

scale_seq

A numeric vector including a set of the maximum spatial distance between a focal individual and neighbors to define neighbor effects. A scalar is also allowed.

addcovar

An optional matrix including additional non-genetic covariates. It contains no. of individuals x no. of covariates.

grouping

A positive integer vector assigning each individual to a group. This argument can be useful when a "smap" contains different experimental replicates. Default setting means that all individuals are belong to a single group.

response

An option to select if the phenotype is a "quantitative" trait subject to linear models, or a "binary" trait subject to logistic models.

n_core

No. of cores for a multi-core computation. This does not work for Windows OS. Default is a single-core computation.

Details

This function uses mixed models via the gaston package (Perdry & Dandine-Roulland 2020). If "binary" is selected, logistic.mm.aireml() is called via the gaston package. In such a case, PVEnei below is replaced by the ratio of phenotypic variation explained (RVE) by neighbor effects as RVE_nei = \sigma_2^2/\sigma_1^2 and p-values are not provided.

Value

A numeric matrix including a given spatial scale, PVE by neighbor effects, and p-values.

• scale Maximum neighbor distance given as an argument

• PVEself Proportion of phenotypic variation explained (PVE) by self effects. RVE is returned when response = "binary"

• PVEnei Proportion of phenotypic variation explained (PVE) by neighbor effects. RVE is returned when response = "binary"

• p-value p-value by a likelihood ratio test between models with or without neighbor effects (when s is not zero); or between a null model and model with self effects alone (when s = 0). NA when response = "binary"

Author(s)

Yasuhiro Sato ([e-mail])

References

Perdry H, Dandine-Roulland C. (2020) gaston: Genetic Data Handling (QC, GRM, LD, PCA) & Linear Mixed Models. https://CRAN.R-project.org/package=gaston

Examples

set.seed(1)
g <- matrix(sample(c(-1,1),100*1000,replace = TRUE),100,1000)
gmap <- cbind(c(rep(1,nrow(g)/2),rep(2,nrow(g)/2)),c(1:ncol(g)))
x <- runif(nrow(g),1,100)
y <- runif(nrow(g),1,100)
smap <- cbind(x,y)
grouping <- c(rep(1,nrow(g)/2),rep(2,nrow(g)/2))
pheno <- nei_simu(geno=g,smap=smap,scale=44,grouping=grouping,n_causal=50,pveB=0.4,pve=0.8)

fake_nei <- list()
fake_nei[[1]] <- g
fake_nei[[2]] <- gmap
fake_nei[[3]] <- smap
fake_nei[[4]] <- data.frame(pheno,grouping)
names(fake_nei) <- c("geno","gmap","smap","pheno")

min_s <- min_dist(fake_nei$smap, fake_nei$pheno$grouping)
scale_seq <- c(min_s, quantile(dist(fake_nei$smap),c(0.2*rep(1:5))))

pve_out <- calc_PVEnei(geno=fake_nei$geno, pheno=fake_nei$pheno[,1],
                       smap=fake_nei$smap, scale_seq=scale_seq,
                       addcovar=as.matrix(fake_nei$pheno$grouping),
                       grouping=fake_nei$pheno$grouping)
delta_PVE(pve_out)