affiXcanTrain: Train the model needed to impute a GReX for each gene

In alussana/AffiXcan: A Functional Approach To Impute Genetically Regulated Expression

Description

Train the model needed to impute a GReX for each gene

Usage

affiXcanTrain(
  exprMatrix,
  assay,
  tbaPaths,
  regionAssoc,
  cov = NULL,
  varExplained = 80,
  scale = TRUE,
  BPPARAM = bpparam(),
  kfold = 1
)

Arguments

exprMatrix	A SummarizedExperiment object containing expression data
assay	A string with the name of the object in SummarizedExperiment::assays(exprMatrix) that contains expression values
tbaPaths	A vector of strings, which are the paths to MultiAssayExperiment RDS files containing the tba values
regionAssoc	A data.frame with the association between regulatory regions and expressed genes and with colnames = c("REGULATORY_REGION", "EXPRESSED_REGION")
cov	Optional. A data.frame with covariates values for the population structure where the columns are the PCs and the rows are the individual IIDs Default is NULL
varExplained	An integer between 0 and 100; varExplained=80 means that the principal components selected to fit the models must explain at least 80 percent of variation of TBA values; default is 80
scale	A logical; if scale=FALSE the TBA values will be only centered, not scaled before performing PCA; default is TRUE
BPPARAM	A BiocParallelParam object. Default is bpparam(). For details on BiocParallelParam virtual base class see browseVignettes("BiocParallel")
kfold	An integer. The k definition of k-fold cross-validation on the training dataset. Default is 1. This argument controls the behavior of the function in the following way: kfold<2: train the models on the whole dataset, not performing the cross-validation kfold>=2: perform the k-fold cross-validation

Value

The output depends on the parameter kfold

If kfold<2: a list containing three objects: pca, bs, regionsCount

• pca: A list containing lists named as the MultiAssayExperiment::experiments() found in the MultiAssayExperiment objects listed in the param tbaPaths. Each of these lists contain two objects:

  • eigenvectors: A matrix containing eigenvectors for those principal components of the TBA selected according to the param varExplained

  • pcs: A matrix containing the principal components values of the TBA selected according to the param varExplained

  • eigenvalues: A vector containing eigenvalues for those principal components of the TBA selected according to the param varExplained

• bs: A list containing lists named as the EXPRESSED_REGIONS found in the param regionAssoc that have a correspondent rowname in the expression values stored SummarizedExperiment::assays(exprMatrix)$assay. Each of the lists in bs contains three objects:

  • coefficients: The coefficients of the principal components used in the model, completely similar to the "coefficients" from the results of lm()

  • p.val: The uncorrected anova pvalue of the model

  • r.sq: The coefficient of determination between the real total expression values and the imputed GReX, retrived from summary(model)$r.squared

  • corrected.p.val: The p-value of the model, corrected for multiple testing with benjamini-hochberg procedure

• regionsCount: An integer, that is the number of genomic regions taken into account during the training phase

If kfold>=2: a list containing k-fold objects, named from 1 to kfold and corresponding to the different cross-validations [i]; each one of these objects is a list containing lists named as the expressed gene IDs [y] (i.e. the rownames() of the object in SummarizedExperiment::assays(exprMatrix) containing the expression values), for which a GReX could be imputed. Each of these inner lists contain two objects:

• rho: the pearson's correlation coefficient (R) between the real expression values and the imputed GReX for the cross-validation i on the expressed gene y, computed with cor()

• rho.sq: the coefficient of determination (R^2) between the real expression values and the imputed GReX for the cross-validation i on the expressed gene y, computed as pearson^2

• cor.test.p.val: the p-value of the cor.test() between the real expression values and the imputed GReX for the cross-validation i on the expressed gene y

• model.p.val: The uncorrected anova pvalue of the model

• model.corrected.p.val: The p-value of the model, corrected for multiple testing with benjamini-hochberg procedure

• model.r.sq: the model's coefficient of determination (R^2) on the training data

Examples

if(interactive()) {
trainingTbaPaths <- system.file("extdata","training.tba.toydata.rds",
package="AffiXcan")

data(exprMatrix)
data(regionAssoc)
data(trainingCovariates)

assay <- "values"

training <- affiXcanTrain(exprMatrix=exprMatrix, assay=assay,
tbaPaths=trainingTbaPaths, regionAssoc=regionAssoc, cov=trainingCovariates,
varExplained=80, scale=TRUE, kfold=3)
}

alussana/AffiXcan documentation built on Oct. 20, 2020, 12:08 a.m.