methylLearn: Machine learning feature selection
In ben-laufer/DMRichR: Enrich your DMR Analysis with the Tidyverse

methylLearn

R Documentation

Machine learning feature selection

Description

Performs feature selection on significant DMRs (predictors) based on random forest (RF) and support vector machine (SVM) algorithms to generate two lists of DMRs ranked by order of importance. Then finds and annotates DMRs that are common among the top percent (or top 10 or number of predictors if top percent is too low) of DMRs in the two DMR ranking lists.

Usage

methylLearn(
  bs.filtered.bsseq = bs.filtered.bsseq,
  sigRegions = sigRegions,
  testCovariate = testCovariate,
  TxDb = NA,
  annoDb = NA,
  topPercent = 1,
  output = "all",
  saveHtmlReport = TRUE
)

Arguments

`bs.filtered.bsseq`	Smoothed `bsseq` object.
`sigRegions`	`GRanges` object of DMRs.
`testCovariate`	The factor to test for differences between groups.
`TxDb`	TxDb annotation package for genome of interest.
`annoDb`	Character specifying OrgDb annotation package for species of interest.
`topPercent`	Positive integer specifying the top percent of DMRs. Default is 1.
`output`	Either "all" or "one". Default is "all". If "output" is "all", then returned object is a list containing tibbles of: 1. full RF variable importance ranking, 2. full SVM variable importance ranking, 3. annotated DMRs common among the top percent (or top 10 or number of predictors if top percent is too low) of DMRs in the two DMR ranking lists. If "output" is "one", then returned object is a tibble of the annotated common DMRs.
`saveHtmlReport`	Either TRUE or FALSE. Default is TRUE. If TRUE, an HTML report with the following is generated: 1. Table of annotated top DMRs from RF DMR importance ranking 2. Table of annotated top DMRs from SVM DMR importance ranking 3. Table of annotated common DMRs 4. Heatmap of each sample of each common DMR If FALSE, no HTML report is generated.