appraise: appraisal for eQTL prediction models
In GGtools: software and data for analyses in genetics of gene expression

Description Usage Arguments Details Value Author(s)

appraisal for eQTL prediction models

appraise(dtab, 
  discretize = TRUE, 
  reduceToSNP = TRUE, 
  prefix, 
  folder = paste0(prefix, "_APPROUT"), 
  discfmlas_in = GGtools:::.discfmlas.demo, 
txlist = list(
    distcats = function(x) {
     cut(x$mindist, c(-1, seq(0, 200001, 50000)))
     },
    fdrcats = function(x) {
     fdrfac = cut(x$fdr, c(-.01, .05, .1, .25, .5, 1.01))
     relevel(fdrfac, "(0.5,1.01]")
     },
    mafcats = function(x) {
     maffac = cut(x$MAF,c(-0.01,.05, .1, .25, .51))
     relevel(maffac, "(-0.01,0.05]")
     },
    caddcats = function(x){
     cut(x$PHRED, c(-.01, 5, seq(10, 30, 10 ), 60))
    }
    ),
    cutts = c(-0.01,seq(0.015,.12,.015),.15), 
    names2check= GGtools:::.standardNames, maxit=30,
     savePinfer=FALSE)
# bindgwava( gwavadt, eqdt )

`dtab`	data.table instance as created by transforming cisRun to GRanges and then to data.table, and then adding CADD PHRED scores if available. If CADD PHRED scores are not available, the default formulas should not be used.
`discretize`	logical telling whether binning to factors defined in txlist should be performed
`reduceToSNP`	logical telling whether ranges should be reduced to unique SNP and FDR recomputed
`prefix`	character atom used to prefix objects saved and folder for result objects
`folder`	folder name suffix
`discfmlas_in`	named list of model formulae
`txlist`	named list of functions that are used to bin certain quantitative features of SNP
`cutts`	numeric vector of thresholds for tabulation and discrete calibration
`names2check`	if NULL, ignored; if a character vector, function will fail unless `all(names2check %in% names(dtab)`
`maxit`	numeric passed to bigglm as control parameter for maximum number of iterations to use in modeling gwas hit probabilities
`savePinfer`	logical specifying whether the inferred probabilities of GWAS involvement are retained

The appraise function wraps many tasks used to appraise eQTL collections in terms of predictive capacity. Details will be provided.

A folder is opened and objects are written representing the test set (data.table on SNPs on even chromosomes), the coefficients of predictive models built on training set (SNPs on odd chromosomes), coefficients of linear regressions of binary test outcomes for calibrating the model on test data, and ROC AUC measures.

bindgwava uses simple data.table operations with match to add three columns to eqdt, gwava_tss, gwava_unmat, and gwava_regi

VJ Carey <stvjc@channing.harvard.edu>

GGtools documentation built on Nov. 8, 2020, 6:32 p.m.