FaST-LMM-EWASher v1.0 README 2/14/2014
From Microsoft Research Factored Spectrally Transformed Linear Mixed Models - FaSTLMM
R/ R code for FaST-LMM-EWASher demo/ Example dataset README-R.txt This document
usage: FLE-script.r data_file phenotype_file [-covar covar_file] [-map map_file]
To run FaST-LMM-EWASher on the example dataset, run this command from the command line in directory .\demo: rscript ..\R\FLE-script.r input_data.txt input_phenotype.txt -covar input_covariates.txt
To optionally specify the map file (see "Inputs" documentation below), use: rscript ..\R\FLE-script.r input_data.txt input_phenotype.txt -covar input_covariates.txt -map input_testmarkers.map
data_file A tab-delimited table of methylation values. In 27k and 450k arrays, this correspond to the beta values of probes. Each column corresponds to a sample and each row corresponds to a probe. The first row is a header: "ID", sampleID1, sampleID2, ... Each subsequent row has the form: probeID, value1, value2, ...
phenotype_file Each row has the form: sampleID sampleStatus. Sample status is 1 if the sample is a case and is 0 otherwise. This file is tab-delimited.
covar_file [optional] The user may specify a set of covariates to include in the model (e.g. age, gender). Each row corresponds to a sample. The order of the samples must be the same as in the data_file header. The first column is always set to '1', the second column is the sample ID, and each subsequent column correspond to a covariate. Do not add column header. This file is tab-delimited.
map_file [optional] This program requires the chromosome and genetic distance of every marker. By default, it assumes a 450K/27K Illumina methylation chip, in which case this parameter can be left blank. If the default is not applicable, please provide a map file, which contains the [chrm#,name,genetic_distance] of each marker, oner per row, with no header, and optionally, extra columns to be ignored. An example is provided in the demo directory (testmarkers.map).
FaST-LMM-EWASher creates two output folders, results/ and tmp/.
The folder results/ contains the following files: out_ewasher.txt. The EWAS P values and statistics for each loci as computed by FaST-LMM-EWASher. FaST-LMM-EWASher filters out loci that are constitutively on or off, and the association statistics are computed for the remaining loci. out_linreg.txt. The P values and statistics from a linear regression with the covariates specified by user. qq_ewasher.png. QQ-plot of the FaST-LMM-EWASher P values. qq_linreg.png. QQ-plot of the linear regression P values. similarity.txt. The similarity matrix computed by FaST-LMM-EWASher. summary.txt. A summary file that states how many loci where analyzed and how many principle components were used by FaST-LMM-EWASher.
The folder tmp/ contains intermediate files produced by FaST-LMM-EWASher.
The file tmp/log.txt contains more detailed run time reports.
FaST-LMM-EWASher uses Intel's MKL library to perform math operations. The program gives an error message if the MKL on the computer is outdated. If this occurs, set the environment variable FastLmmUseAnyMklLib=1. For example, in Unix, add "export FastLmmUseAnyMklLib=1" in the .bashrc or .cshrc file.
By default, this software uses REML to estimate the linear mixed model parameters, in conjunction with an F-test to obtain p-values.
There are a few parameters currently hard-coded, at the top of fastlmm-ewasher.py and can be changed there: #constitutively unmethylated/methylated filter values minFilterVal = 0.2; maxFilterVal = 0.8
#values for fastlmmc autoselect to try for number of loci in GSM #may be advantageous to try a finer, or more enlarged grid, but this default setting seems reasonable. asVals = append(append(c(1), seq(100, 900, 100)), seq(1000, 10000, 1000))
maxPC = 10
#threshold on lambda for when to stop adding PCs lamdaThresh = 1.0