curveFDP {curveFDP}R Documentation

Fitting of a two- or three component mixture model to obtain False Discovery Proportion

Description

function to estimate global fdr by curve-fitting a gaussian mixture model (gmm) of two or three distributions, one, to the left, for spectra of evident very low quality (if necessary), one, in the center, for spectra with random hits to sequences one, to the right, with non-random hits to sequences the three distributions may overlap, the gmm is used to find the best seperation assumption is that quantile of the random-hit distribution can be used to estimate how likely a score is to stem from that distribution and thus to be random.

Usage

curveFDP(scores, FDPperc=5, bootstrap=1)

Arguments

scores scores corresponds to the results of the scoring of sequences to spectra (vector), for each spectrum only the best score, higher scores are better
FDPperc FDPperc is the desired FDP cutoff in percent (so 5 for 0.05 likelihood of having a value that extreme or higher by random)
bootstrap The number of times the FDP computation should be repeated with bootstrap resampling of datasets of the same size

Details

function to estimate global fdr by curve-fitting a gaussian mixture model (gmm) of two or three distributions, one, to the left, for spectra of evident very low quality (if necessary), one, in the center, for spectra with random hits to sequences one, to the right, with non-random hits to sequences the three distributions may overlap, the gmm is used to find the best seperation assumption is that quantile of the random-hit distribution can be used to estimate how likely a score is to stem from that distribution and thus to be random.

Value

FDPcutoff the score atrting at which there is a likelihood of FDPperc or less that a hit is random
aboveFDR number of scores above the FDPcutoff
pvalues pvalues from the chi square test for appropriateness of the mixture model
out mixture model as identified by mixtools package
modelsize number of components in the mixture model

Author(s)

Bernhard Renard bernhard.renard@iwr.uni-heidleberg.de


[Package curveFDP version 2.0 Index]