| Title: | Latent Interaction Testing for Genome-Wide Studies |
|---|---|
| Description: | Identifying latent genetic interactions in genome-wide association studies using the Latent Interaction Testing (LIT) framework. LIT is a flexible kernel-based approach that leverages information across multiple traits to detect latent genetic interactions without specifying or observing the interacting variable (e.g., environment). LIT accepts standard PLINK files as inputs to analyze large genome-wide association studies. |
| Authors: | Andrew Bass [aut, cre], Michael Epstein [aut] |
| Maintainer: | Andrew Bass <[email protected]> |
| License: | LGPL |
| Version: | 1.0.0 |
| Built: | 2024-09-09 04:42:40 UTC |
| Source: | https://github.com/ajbass/lit |
The GAMuT function is a kernel-based multivariate association test.
Note that our code to process plink files builds from the
genio R package.
gamut_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)gamut_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)
y |
matrix of traits (n observations by k traits) |
file |
path to plink files |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
verbose |
If TRUE (default) print progress. |
A data frame of p-values where the columns are the cross products/squared residuals and the rows are SNPs.
# set seed set.seed(123) # path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- gamut_plink(Y, file = file)# set seed set.seed(123) # path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- gamut_plink(Y, file = file)
lit performs a kernel-based testing procedure, Latent Interaction Testing (LIT), using a set of traits and SNPs.
LIT tests whether the squared residuals (SQ) and cross products (CP) are statistically independent of the genotypes.
In particular, we construct a kernel matrix for the SQ/CP terms to measure the pairwise
similarity between individuals, and also construct an analogous one for the genotypes.
We then test whether these two matrices are independent.
Currently, we implement the linear and projection kernel functions to measure pairwise similarity between individuals.
We then combine the p-values of these implementations using a Cauchy combination test to maximize the number of discoveries.
lit(y, x, adjustment = NULL, pop_struct = NULL)lit(y, x, adjustment = NULL, pop_struct = NULL)
y |
matrix of traits (n observations by k traits) |
x |
matrix of SNPs (n observations by m SNPs) |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
A data frame of p-values where the columns are
wlit: LIT using a linear kernel
ulit: LIT using a projection kernel
alit: Cauchy combination test of the above two LIT implementations.
# set seed set.seed(123) # Generate SNPs and traits X <- matrix(rbinom(10*2, size = 2, prob = 0.25), ncol = 2) Y <- matrix(rnorm(10*4), ncol = 4) out <- lit(Y, X)# set seed set.seed(123) # Generate SNPs and traits X <- matrix(rbinom(10*2, size = 2, prob = 0.25), ncol = 2) Y <- matrix(rnorm(10*4), ncol = 4) out <- lit(Y, X)
Internal use for now
lit_h(y, x, adjustment = NULL, pop_struct = NULL)lit_h(y, x, adjustment = NULL, pop_struct = NULL)
y |
matrix of traits (n observations by k traits) |
x |
matrix of SNPs (n observations by m SNPs) |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
lit_plink performs a kernel-based testing procedure, Latent Interaction Testing (LIT), using a set of traits and SNPs.
LIT tests whether the squared residuals (SQ) and cross products (CP) are statistically independent of the genotypes.
In particular, we construct a kernel matrix for the SQ/CP terms to measure the pairwise
similarity between individuals, and also construct an analogous one for the genotypes.
We then test whether these two matrices are independent.
Currently, we implement the linear and projection kernel functions to measure pairwise similarity between individuals.
We then combine the p-values of these implementations using a Cauchy combination test to maximize the number of discoveries.
This function is suitable for large datasets (e.g., UK Biobank) in plink format.
Note that our code to process plink files builds from the
genio R package
lit_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)lit_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)
y |
matrix of traits (n observations by k traits) |
file |
path to plink files |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
verbose |
If TRUE (default) print progress. |
A data frame of p-values where the columns are
wlit: LIT using a linear kernel
ulit: LIT using a projection kernel
alit: Cauchy combination test of the above two LIT implementations.
# set seed set.seed(123) # path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- lit_plink(Y, file = file)# set seed set.seed(123) # path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- lit_plink(Y, file = file)
The marginal function performs a trait-by-trait univariate test for latent interactions
using the squared residuals and cross products.
marginal(y, x, adjustment = NULL, pop_struct = NULL)marginal(y, x, adjustment = NULL, pop_struct = NULL)
y |
matrix of traits (n observations by k traits) |
x |
matrix of SNPs (n observations by m SNPs) |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
A data frame of p-values where the columns are the cross products/squared residuals and the rows are SNPs.
# set seed set.seed(123) # Generate SNPs and traits X <- matrix(rbinom(10*2, size = 2, prob = 0.25), ncol = 2) Y <- matrix(rnorm(10*4), ncol = 4) out <- marginal(Y, X)# set seed set.seed(123) # Generate SNPs and traits X <- matrix(rbinom(10*2, size = 2, prob = 0.25), ncol = 2) Y <- matrix(rnorm(10*4), ncol = 4) out <- marginal(Y, X)
The marginal_plink function performs a trait-by-trait univariate test for latent interactions
using the squared residuals and cross products. This function is suitable for large
datasets (e.g., UK Biobank) in plink format. Note that our code to process plink files builds from the
genio R package.
marginal_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)marginal_plink(y, file, adjustment = NULL, pop_struct = NULL, verbose = TRUE)
y |
matrix of traits (n observations by k traits) |
file |
path to plink files |
adjustment |
matrix of covariates to adjust traits |
pop_struct |
matrix of PCs that captures population structure |
verbose |
If TRUE (default) print progress. |
A data frame of p-values where the columns are the cross products/squared residuals and the rows are SNPs.
# set seed set.seed(123) # Path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- marginal_plink(Y, file = file)# set seed set.seed(123) # Path to plink files file <- system.file("extdata", 'sample.bed', package = "genio", mustWork = TRUE) # Generate trait expression Y <- matrix(rnorm(10*4), ncol = 4) out <- marginal_plink(Y, file = file)