Train and predict gene's predictive model with mediators using MeTWAS

The function trains a predictive model of a given gene using top mediators as fixed effects and assesses in-sample performance with cross-validation.

trainExpression(
  geneInt,
  snps,
  snpLocs,
  mediator,
  medLocs,
  covariates,
  dimNumeric,
  qtlFull,
  h2Pcutoff = 0.1,
  numMed = 5,
  seed,
  k,
  cisDist = 5e+05,
  parallel = T,
  prune = F,
  windowSize = 50,
  numSNPShift = 5,
  ldThresh = 0.5,
  cores,
  verbose = T,
  LDMS = F,
  modelDir,
  ldScrRegion = 200,
  snpAnnot = NULL
)

Arguments

geneInt	character, identifier for gene of interest
snps	data frame, SNP dosages
snpLocs	data frame, MatrixEQTL locations for SNPs
mediator	data frame, mediator intensities
medLocs	data frame, MatrixEQTL locations for mediators
covariates	data frame, covariates
qtlFull	data frame, all QTLs (cis and trans) between mediators and genes
h2Pcutoff	numeric, P-value cutoff for heritability
numMed	integer, number of top mediators to include
seed	integer, random seed for splitting
k	integer, number of training-test splits
parallel	logical, TRUE/FALSE to run glmnet in parallel
prune	logical, TRUE/FALSE to LD prune the genotypes
windowSize	integer, window size for PLINK pruning
numSNPShift	integer, shifting window for PLINK pruning
ldThresh	numeric, LD threshold for PLINK pruning
cores	integer, number of parallel cores
outputAll	logical, include mediator information

Value

final model for gene along with CV R2 and predicted values