msqrob: Methods to fit msqrob models with ridge regression and/or...

msqrob,SummarizedExperiment-methodR Documentation

Methods to fit msqrob models with ridge regression and/or random effects using lme4

Description

Parameter estimation of msqrob models for QFeatures and SummarizedExperiment instance.

Usage

## S4 method for signature 'SummarizedExperiment'
msqrob(
  object,
  formula,
  modelColumnName = "msqrobModels",
  overwrite = FALSE,
  robust = TRUE,
  ridge = FALSE,
  maxitRob = 1,
  tol = 1e-06,
  doQR = TRUE,
  lmerArgs = list(control = lmerControl(calc.derivs = FALSE))
)

## S4 method for signature 'QFeatures'
msqrob(
  object,
  i,
  formula,
  modelColumnName = "msqrobModels",
  overwrite = FALSE,
  robust = TRUE,
  ridge = FALSE,
  maxitRob = 1,
  tol = 1e-06,
  doQR = TRUE,
  lmerArgs = list(control = lmerControl(calc.derivs = FALSE))
)

Arguments

object

SummarizedExperiment or QFeatures instance

formula

Model formula. The model is built based on the covariates in the data object.

modelColumnName

character to indicate the variable name that is used to store the msqrob models in the rowData of the SummarizedExperiment instance or of the assay of the QFeatures instance. Default is "msqrobModels".

overwrite

boolean(1) to indicate if the column in the rowData has to be overwritten if the modelColumnName already exists. Default is FALSE.

robust

boolean(1) to indicate if robust regression is performed to account for outliers. Default is TRUE. If FALSE an OLS fit is performed.

ridge

boolean(1) to indicate if ridge regression is performed. Default is FALSE. If TRUE the fixed effects are estimated via penalized regression and shrunken to zero.

maxitRob

numeric(1) indicating the maximum iterations in the IRWLS algorithm used in the M-estimation step of the robust regression.

tol

numeric(1) indicating the tolerance for declaring convergence of the M-estimation loop.

doQR

boolean(1) to indicate if QR decomposition is used when adopting ridge regression. Default is TRUE. If FALSE the predictors of the fixed effects are not transformed, and the degree of shrinkage can depend on the encoding.

lmerArgs

a list (of correct class, resulting from ‘lmerControl()’ containing control parameters, including the nonlinear optimizer to be used and parameters to be passed through to the nonlinear optimizer, see the ‘lmerControl’ documentation of the lme4 package for more details. Default is list(control = lmerControl(calc.derivs = FALSE))

i

character or integer to specify the element of the QFeatures that contains the log expression intensities that will be modelled.

Value

A SummarizedExperiment or a QFeatures instance with the models.

Author(s)

Lieven Clement

Examples


# Load example data
# The data are a Feature object with containing
# a SummarizedExperiment named "peptide" with MaxQuant peptide intensities
# The data are a subset of spike-in the human-ecoli study
# The variable condition in the colData of the Feature object
# contains information on the spike in condition a-e (from low to high)
data(pe)

# Aggregate peptide intensities in protein expression values
pe <- aggregateFeatures(pe, i = "peptide", fcol = "Proteins", name = "protein")

# Fit MSqrob model using robust linear regression upon summarization of
# peptide intensities into protein expression values.
# For summarized SummarizedExperiment
se <- pe[["protein"]]
se
colData(se) <- colData(pe)
se <- msqrob(se, formula = ~condition, modelColumnName = "rlm")
getCoef(rowData(se)$rlm[[1]])

# For features object
pe <- msqrob(pe, i = "protein", formula = ~condition, modelColumnName = "rlm")
# with ridge regression (slower)
pe <- msqrob(pe, i = "protein", formula = ~condition, ridge = TRUE, modelColumnName = "ridge")

# compare for human protein (no DE)==> large shrinkage to zero
cbind(getCoef(rowData(pe[["protein"]])$rlm[[1]]), getCoef(rowData(pe[["protein"]])$ridge[[1]]))

# compare for ecoli protein (DE)==> almost no shrinkage to zero
cbind(
    getCoef(rowData(pe[["protein"]])$rlm[["P00956"]]),
    getCoef(rowData(pe[["protein"]])$ridge[["P00956"]])
)

statOmics/msqrob2 documentation built on May 8, 2024, 4:38 p.m.