GWAS {BGData}

Performs Single Marker Regressions Using BGData Objects

Description

Implements single marker regressions. The regression model includes all the covariates specified in the right-hand-side of the formula plus one column of the genotypes at a time. The data from the association tests is obtained from a BGData object.

Usage

GWAS(formula, data, method = "lsfit", i = seq_len(nrow(geno(data))),
  j = seq_len(ncol(geno(data))), chunkSize = 5000L,
  nCores = getOption("mc.cores", 2L), verbose = FALSE, ...)

Arguments

formula

The formula for the GWAS model without the variant, e.g. y ~ 1 or y ~ factor(sex) + age. The variables included in the formula must be column names in the sample information of the BGData object.

data

A BGData object.

method

The regression method to be used. Currently, the following methods are implemented: rayOLS (see below), lsfit, lm, lm.fit, glm, lmer, and SKAT. Defaults to lsfit.

i

Indicates which rows of the genotypes should be used. Can be integer, boolean, or character. By default, all rows are used.

j

Indicates which columns of the genotypes should be used. Can be integer, boolean, or character. By default, all columns are used.

chunkSize

The number of columns of the genotypes that are brought into physical memory for processing per core. If NULL, all elements in j are used. Defaults to 5000.

nCores

The number of cores (passed to mclapply). Defaults to the number of cores as detected by detectCores.

verbose

Whether progress updates will be posted. Defaults to FALSE.

...

Additional arguments for chunkedApply and regression method.

Details

The rayOLS method is a regression through the origin that can only be used with a y ~ 1 formula, i.e. it only allows for one quantitative response variable y and one variant at a time as an explanatory variable (the variant is not included in the formula, hence 1 is used as a dummy). If covariates are needed, consider preadjustment of y. Among the provided methods, it is by far the fastest.

Some regression methods may require the data to not contain columns with variance 0 or too many missing values. We suggest running summarize to detect variants that do not clear the desired minor-allele frequency and rate of missing genotype calls, and filtering these variants out using the j parameter of the GWAS function (see example below).

Value

The same matrix that would be returned by coef(summary(model)).

See Also

file-backed-matrices for more information on file-backed matrices. multi-level-parallelism for more information on multi-level parallelism. BGData-class for more information on the BGData class. lsfit, lm, lm.fit, glm, lmer, and SKAT for more information on regression methods.

Examples

# Restrict number of cores to 1 on Windows
if (.Platform$OS.type == "windows") {
    options(mc.cores = 1)
}

# Load example data
bg <- BGData:::loadExample()

# Detect variants that do not pass MAF and missingness thresholds
summaries <- summarize(geno(bg))
maf <- ifelse(summaries$allele_freq > 0.5, 1 - summaries$allele_freq,
    summaries$allele_freq)
exclusions <- maf < 0.01 | summaries$freq_na > 0.05

# Perform a single marker regression
res1 <- GWAS(formula = FT10 ~ 1, data = bg, j = !exclusions)

# Draw a Manhattan plot
plot(-log10(res1[, 4]))

# Use lm instead of lsfit (the default)
res2 <- GWAS(formula = FT10 ~ 1, data = bg, method = "lm", j = !exclusions)

# Use glm instead of lsfit (the default)
y <- pheno(bg)$FT10
pheno(bg)$FT10.01 <- y > quantile(y, 0.8, na.rm = TRUE)
res3 <- GWAS(formula = FT10.01 ~ 1, data = bg, method = "glm", j = !exclusions)

# Perform a single marker regression on the first 50 markers (useful for
# distributed computing)
res4 <- GWAS(formula = FT10 ~ 1, data = bg, j = 1:50)