wrapMinfi: Finding differentially methylated CpG positions or regions

In frankRuehle/systemsbio: Streamlined Analysis and Integration of Systems Biology Data

Description

Use minfi-package to determine either differentially methylated positions (DMPs, dmpFinder) and/or differentially methylated regions (DMRs, bumphunter).

Usage

wrapMinfi(
  GRset,
  projectfolder = "MT/minfi",
  projectname = NULL,
  phenotype,
  type.covar = "categorical",
  returnResults = c("DMP", "DMR"),
  qCutoff = 0.001,
  bumpcutoff = 0.1,
  nResamples = 1000,
  maxGap = 500
)

Arguments

GRset	GenomicRatioSet or MethylSet
projectfolder	character with directory for output files (will be generated if not exisiting).
projectname	optional character prefix for output file names.
phenotype	character with phenotype for differential methylation analysis. Must be given within the phenotype data of GRset.
type.covar	character with type of phenotype variable, either continuous or categorical.
returnResults	character indicating differentially methylated elements to return. Either "DMP" for differentially methylated positions using dmpFinder or "DMR" for differentially methylated regions using bumphunter or both.
qCutoff	numeric with q-value (FDR) threshold for DMPs to be reported by dmpFinder.
bumpcutoff	numeric start value to find the appropriate bumphunter cutoff. A cutoff 0.1 corresponds to 10 If the cutoff results in more than 10000 candidate DMRs, the cutoff is automatically increased before applying permutation testing to avoid excessive computation time.
nResamples	numeric, number of resamples to use when computing null distributions with bumphunter.
maxGap	numeric with maximum location gap used to define clusters of probes for bumphunter.

Details

The dmpFinder function from minfi identifies differentially methylated CpG-sites using linear regression for continuous phenotypes and F-test for categorical phenotypes. In case of a categorical phenotype with many groups (e.g. experimental groups), F-test is applied over all groups. Tests are performed on M values, which are logit transformed Beta values (beta = Methylated allele intensity / (Unmethylated allele intensity + Methylated allele intensity + 100)). Annotation data of GRset is added to the result table.

Instead of looking for association between a single genomic location and a phenotype of interest, bumphunter looks for genomic regions that are differentially methylated (beta values). In the context of the 450k array, the algorithm first defines clusters of probes such that two consecutive probe locations in the cluster are not separated by more than distance mapGap. Briefly, the algorithm first computes a t-statistic for beta values at each genomic location, with optional smoothing. Then, it defines a candidate region to be a cluster of probes for which all the t-statistics exceed a predefined threshold. To test for significance of the candidate regions, the algorithm uses permutations given in nResamples.

Value

list with up to 2 elements depending on selection in returnResults. Result tables are also stored as side-effects in projectfolder

• DMP dataframe of CpG positions sorted by differential methylation p-value annotated by respective annotation package.

• DMR An object of class bumps with the components: tab, coef, fitted, pvaluesMarginal, null, algorithm. See help(bumphunter) for details. The dataframe in tab is annotated with meta data for all CpG-sites residing within the respective DMR.

Author(s)

Frank Ruehle

frankRuehle/systemsbio documentation built on Sept. 14, 2020, 1:18 a.m.