getDataByFilter: Return the data by feature filtering

In transbioZI/BioMMex: BioMM: Biological-informed Multi-stage Machine learning framework for phenotype prediction using omics data

Return the data by feature filtering

Description

Identify and select a subset of outcome-associated or predictive features in the training data based on filtering methods. Return the same set of selected features for the test data if it is available.

Usage

getDataByFilter(
  trainData,
  testData,
  FSmethod,
  cutP = 0.1,
  fdr = NULL,
  FScore = MulticoreParam()
)

Arguments

trainData	The input training dataset. The first column is the label.
testData	The input test dataset. The first column is the label.
FSmethod	Feature selection methods. Available options are c(NULL, 'positive', 'wilcox.test', 'cor.test', 'chisq.test', 'posWilcox', or 'top10pCor'). 'positive' is the positively outcome-associated features using the Pearson correlation method. 'posWilcox' is the positively outcome-associated features using Pearson correlation method together with 'wilcox.text' method. 'top10pCor' is the top 10 outcome-associcated features. This is helpful when no features can be picked during stringent feature selection procedure. 'posTopCor' selects the number of most correlated features.
cutP	The cutoff used for p value thresholding. It can be any value between 0 and 1. Commonly used cutoffs are c(0.5, 0.1, 0.05, 0.01, etc.). The default is 0.1. If FSmethod = "posTopCor", cutP is then defined as the number of most correlated features with 'fdr' = NULL.
fdr	Multiple testing correction method. Available options are c(NULL, 'fdr', 'BH', 'holm' etc). See also p.adjust. The default is NULL.
FScore	The number of cores used for some feature selection methods. If it's NULL, then no parallel computing is applied.

Details

Parallel computing is helpful if your input data is high dimensional. For 'cutP', a soft thresholding of 0.1 may be favorable than more stringent p value cutoff because the features with small effect size can be taken into consideration for downstream analysis. However, for high dimensional (e.g. p > 10,000) data, many false positive features may exist, thus, rigorous p value thresholding should be applied. The choice of feature selection method depends on the characteristics of the input data.

Value

Both training and test data (if provided) with pre-selected features are returned if feature selection method is applied. If no feature can be selected during feature selection procedure, then the output is NULL.

Author(s)

Junfang Chen

Examples

 
## Load data  
methylfile <- system.file('extdata', 'methylData.rds', package='BioMM')  
methylData <- readRDS(methylfile)  
trainIndex <- sample(nrow(methylData), 20)
trainData = methylData[trainIndex,]
testData = methylData[-trainIndex,]
## Feature selection
library(BiocParallel)
param <- MulticoreParam(workers = 10)
## Select outcome-associated features based on the Wilcoxon test (P<0.1)
datalist <- getDataByFilter(trainData, testData, FSmethod="wilcox.test", 
                           cutP=0.1, fdr=NULL, FScore=param)
trainDataSub <- datalist[[1]] 
testDataSub <- datalist[[2]] 
print(dim(trainData))
print(dim(trainDataSub))

transbioZI/BioMMex documentation built on Jan. 27, 2023, 4:14 a.m.