benchmark_shared_hits: benchmark_shared_hits

In scsR: SiRNA correction for seed mediated off-target effect

Description

This method can be used to benchmark sorted gene vectors (A) that comes out from a siRNA screen. The benchmark is done against other sorted gene vectors (B) that we know to contain high density of real hits (e.g. the results of a second siRNA screen performed with a different library). The benchmark is performed simply comparing the top n hits of the two lists. If the two lists contain many shared best hits than we have a strong statistical signal. Then we display the number of shared best hits for different n, in a graph (if visualize_pval variable is set to true the pvalue of the t-test is plotted instead of the number of shared hits).

Usage

benchmark_shared_hits(glA, glB, col, avoidIntersectL=FALSE, 
                                     output_file=NULL, npoints=400, title="", scaleAXPoint = 1, 
                                     scaleBXPoint = NULL, fixedBXPoint=400, displayRandomMultipleLines=TRUE, 
                                     nrandom=20, intersectGenes=TRUE, visualize_pval=FALSE, max_ylim=NULL, xlab=NULL, ylab="shared hits")

Arguments

glA	sorted list containing one or more sorted vectors of genes (i.e. hits of a genome wide screen sorted by significance). Each element i of glA will be benchmarked against element i of glB. In case glB contains only one element, each glA vector will be benchmarked against glB[1].
glB	sorted list containing one or more sorted vectors of genes (i.e. hits of a genome wide screen sorted by significance).
col	sorted vector of booleans (a boolean i in the vector corresponds to the shared hits of glA[i] with glB[i] )
avoidIntersectL	sorted vector of colors (a color i in the vector corresponds to the shared hits line obtain intersecting glA[i] with glB[i] ) To perform the benchmark we construct a background to be used (this background is given by the intersection of all the glA and glB vectors) When an element i of the vector is set to TRUE, we don't use the elements of glA[i] to compute the vector. This allows to benchmark also methods that do output only few putative good genes (instead of a sorted list of all the genes tested).
npoints	number of points on the x-axis of the graph (integer)
nrandom	number of random lines to compute (in order to infer the variation of the noise) (integer)
output_file	path to the output file where to store the graph (character vector)
title	title of the graph (character vector)
scaleAXPoint	for position x in the graph we compare the best x * scaleAXPoint best hits of the genesA vector (integer)
scaleBXPoint	for position x in the graph we compare the best x * scaleBXPoint best hits of the genesB vector (integer)
fixedBXPoint	for position x in the graph we compare the best fixedBXPoint best hits of the genesB vector (integer)
intersectGenes	specify whether to intersect the genes from the various input vectors to form a suitable background to be used for the benchmark. (boolean)
visualize_pval	specify whether a p-value (derived by an hypergeometric test) should be visualized instead of the number of shared hits. (boolean)
displayRandomMultipleLines	specify whether to display several random lines in the graph (instead of only one line that is the average of all the random lines) (boolean)
max_ylim	y upper limit (integer)
xlab	xlab (character vector)
ylab	ylab (character vector)

Author(s)

Andrea Franceschini

Examples

data(uuk_screen)
data(uuk_screen_dh)

benchmark_shared_hits(
  glA=list(
    uuk_screen[1:1000,]$GeneID, 
    arrange(add_rank_col(uuk_screen[1:1000,]), log_pval_rsa)$GeneID
  ),
  glB=list(uuk_screen_dh$GeneID),
  col=c("black", "blue"),
  title="UUKUNIEMI Hela Cell Killers"
)

scsR documentation built on April 28, 2020, 7:11 p.m.