kmFit: Linear mixed effects models with kinship for RNA-seq

In BIGslu/kimma: Kinship In Mixed Model Analysis of RNA-seq

Linear mixed effects models with kinship for RNA-seq

Description

Run lmerel and corresponding lm or lme without kinship of gene expression in RNA-seq data

Usage

kmFit(
  dat = NULL,
  kin = NULL,
  patientID = "ptID",
  libraryID = "libID",
  counts = NULL,
  meta = NULL,
  genes = NULL,
  weights = NULL,
  subset_var = NULL,
  subset_lvl = NULL,
  subset_genes = NULL,
  model,
  use_weights = FALSE,
  run_lm = FALSE,
  run_lme = FALSE,
  run_lmerel = FALSE,
  metrics = FALSE,
  run_contrast = FALSE,
  contrast_var = NULL,
  processors = NULL,
  p_method = "BH",
  genotype_name = NULL,
  run.lmekin = NULL,
  subset.var = NULL,
  subset.lvl = NULL,
  subset.genes = NULL,
  use.weights = FALSE,
  run.lm = FALSE,
  run.lme = FALSE,
  run.lmerel = FALSE,
  run.contrast = FALSE,
  contrast.var = NULL,
  p.method = NULL
)

Arguments

dat	EList object output by voom( ). Must contain counts (dat$E) and meta (dat$targets). Optionally also contains gene metadata (dat$genes) and weights (dat$weights)
kin	Matrix with pairwise kinship values between individuals. Must be numeric with rownames.
patientID	Character of variable name to match dat$targets to kinship row and column names.
libraryID	Character of variable name to match dat$targets to dat$E colnames
counts	Matrix of normalized expression. Rows are genes, columns are libraries.
meta	Matrix or data frame of sample and individual metadata.
genes	Optional matrix or data frame of gene metadata.
weights	Optional matrix of data frame of gene specific weights. Usually calculated with limma::voomWithQualityWeights().
subset_var	Character list of variable name(s) to filter data by.
subset_lvl	Character list of variable value(s) or level(s) to filter data to. Must match order of subset_var
subset_genes	Character vector of genes to include in models.
model	Character vector of model starting with ~ Should include (1|patientID) if mixed effects will be run
use_weights	Logical if gene specific weights should be used in model. Default is FALSE
run_lm	Logical if should run lm model without kinship
run_lme	Logical if should run lme model without kinship
run_lmerel	Logical if should run lmerel model with kinship
metrics	Logical if should calculate model fit metrics such as AIC, BIC, R-squared. Default is FALSE
run_contrast	Logical if should run pairwise contrasts. If no matrix provided, all possible pairwise comparisons are completed.
contrast_var	Character vector of variable in model to run contrasts of. Interaction terms must be specified as "var1:var2". If NULL (default), all contrasts for all variables in the model are run
processors	Numeric processors to run in parallel. Default is 2 less than the total available
p_method	Character of FDR adjustment method. Values as in p.adjust( )
genotype_name	Character string. Used internally for kmFit_eQTL
run.lmekin	Deprecated. Please use run_lmerel
subset.var	Deprecated form of subset_var
subset.lvl	Deprecated form of subset_lvl
subset.genes	Deprecated form of subset_genes
use.weights	Deprecated form of use_weights
run.lm	Deprecated form of run_lm
run.lme	Deprecated form of run_lme
run.lmerel	Deprecated form of run_lmerel
run.contrast	Deprecated form of run_contrast
contrast.var	Deprecated form of contrast_var
p.method	Deprecated form of p_method

Value

List of data frames including - lm/lme/lmerel: model estimates and significance - *.contrast: model estimates and significance for pairwise contrasts with variables in the original model - *.fit: model fit metrics such as sigma, AIC, BIC, R-squared (optional with metrics paramater) - *.error: error messages for genes that failed model fitting

Examples

# All samples and all genes
## Not run
# kmFit(dat = example.voom,
#     kin = example.kin, run_lmerel = TRUE,
#     model = "~ virus + (1|ptID)")

# Subset samples and genes
## Also with weights
kmFit(dat = example.voom,
      run_lm = TRUE, use_weights = FALSE,
      subset_var = list("asthma"), subset_lvl = list(c("asthma")),
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ virus + (1|ptID)")

# Pairwise contrasts
## Continuous interaction
kmFit(dat = example.voom,
      run_lme = TRUE, run_contrast = TRUE,
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ virus + asthma * median_cv_coverage + (1|ptID)",
      contrast_var=c("asthma:median_cv_coverage"))

## Categorical interaction
kmFit(dat = example.voom, kin = example.kin,
      run_lmerel = TRUE, run_contrast = TRUE, metrics=TRUE,
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ virus*asthma + (1|ptID)",
      contrast_var=c("virus:asthma"))

# Model with failed genes
kmFit(dat = example.voom,
      kin = example.kin, run_lmerel = TRUE, run_lm = TRUE,
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ virus*asthma + lib.size + norm.factors + median_cv_coverage + ptID + (1|ptID)")

# Non-dat data
kmFit(counts = example.voom$E, meta = example.voom$targets,
      run_lm = TRUE, use_weights = FALSE,
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ virus + (1|ptID)")

# Three level variable
example.voom$targets$lvl <- rep(c("A","B","C"), length(example.voom$targets$libID)/3)
kmFit(dat = example.voom,
      run_lme= TRUE, run_contrast = TRUE,
      subset_genes = c("ENSG00000250479","ENSG00000250510","ENSG00000255823"),
      model = "~ lvl + (1|ptID)")

BIGslu/kimma documentation built on May 30, 2024, 10:09 p.m.