matrixEpistasis: matrixEpistasis

In fanglab/MatrixEpistasis: An ultrafast exhaustive epistasis scan for quantitative traits with covariate adjustment

Description

matrixEpistasis uses large matrix operation to perform the exhaustive epistasis scan for quantitative traits with covariate adjustment

Usage

matrixEpistasis(snpA, snpB, trait, covariate = NULL)

Arguments

snpA	a matrix of numeric values in size of sample*snp representing the 1st group of SNPs, where the column names are the snp_ids
snpB	a matrix of numeric values in size of sample*snp representing the 2nd group of SNPs, where the column names are the snp_ids
trait	a vector of numeric values representing the quantitative trait
covariate	a matrix of numeric values in size of sample*covariate, by default, NULL

Value

A list containing the follow components:

• r a matrix of numeric values representing the partial correlation coefficients between snpA*snpB and traits conditioned on snpA, snpB and covariates

• df an integer value representing the degree of freedom

Author(s)

Shijia Zhu, [e-mail]

See Also

p2c; matrixPval; smoothQQplot

Examples

# randomly generate a SNP matrix
snp <- sapply(1:100,function(i) rnorm(1000) )
# assign names to SNPs
colnames(snp) <- paste0('snp',1:100)
snpA = snp
snpB = snp

# radnomly generate a quantitative trait by simulating the relationship between SNPs and traits
trait <- snp %*% rnorm(100)

# use the top 5 PCs as the covariates 
covariate <- prcomp(snp)$x[,1:5]

# run matrixEpistasis with covariates adjustment
res <- matrixEpistasis( snpA=snpA, snpB=snpB, trait=trait, covariate = covariate )
r <- res$r
df <- res$df

# run matrixEpistasis function with covariates adjustment
res <- matrixEpistasis( snpA=snpA, snpB=snpB, trait=trait, covariate = covariate )

# res is a list comprising two components: r and df. res$r is the matrix of partial correlation coefficients between snp interaction (snpA*snpB) and trait with additive effects (snpA and snpB) and covariates adjusted, and res$df is the degree of freedom for the partial correlation.
names(res)
r = res$r 
df = res$df

# based on the degree of freedom, run matrixPval function to calculate p values for all partial correlation coefficients. The result is a matrix of p-values for epistasis of all snp interactions.
p <- matrixPval( r , df )

# alternatively, users can calculate p-values only for those entries with p vlaues less than a given threshold, say 1e-2, shown as follows: 
# use p2c to covert p value threshold 1e-5 to the corresponding partial correlation coefficient
corrThreshold <- p2c( pval=1e-2 , df )

# extract the index for those significant ones
index <- which( abs(r)>corrThreshold , arr.ind=TRUE )

# get the SNP names
snp1 <- colnames(snpA)[ index[,1] ]
snp2 <- colnames(snpB)[ index[,2] ]

# use matrixPval function to calculate p values for only those of interest
pvalue <- matrixPval( r[index] , df  )

# build the data frame
sig_res <- data.frame( snp1 , snp2 , pvalue )
head(sig_res)

fanglab/MatrixEpistasis documentation built on May 25, 2019, 5:22 p.m.