calcHypPI: Probability intervals calculation for CAT curves using the...

Description Usage Arguments Details Value Note Author(s) References See Also Examples

Description

The calcHypPI function calculates probability intervals for a correspondence at the top (CAT) curve using the hypergeometric distribution. This function, based on the qhyper quantile function, produces a probability intervals matrix to be passed as argument to plotCat in order to add probability intervals shades when plotting CAT curves.

Usage

1
calcHypPI(data,  expectedProp = 0.1, prob = c(0.999999,0.999,0.99,0.95))

Arguments

data

The same data frame used to compute the CAT curves with the computeCat function. It contains a column of unique identifiers and at least two columns of ranking statistics.

expectedProp

A single numeric value between 0 and 1. This is the proportion of features expected to be corresponding at the top of the ranking. The "expectedProp" argument can be set to NULL if the number of features expected to be similarly ranked is unknown.

prob

A numeric vector specifying the probabiliy intervals for the CAT curves to be computed.

Details

The calcHypPI uses qhyper quantile function to compute the proportions of common features between two ordered vectors for specified quantiles of the hypergeometric distribution. Such proportions are used to add probability intervals to CAT curves computed using ranks (see computeCat). The prob argument is used to specify the desired probability intervals to be computed. By default this numeric vector is equal to c(0.999999, 0.999, 0.99, 0.95).

To understand the way this function works we can use the analogy of repeated drawing of an increasing number of balls from an urn containing both white and black balls (see qhyper). According to this analogy the total number of balls in the urn corresponds to the total number of common features between two ordered vectors that are being compared (e.g. all the genes in common between two genomic studies).

The number of white balls corresponds to the top ranking features that are correctly ordered (successes), while the black balls represent the features that are not correctly ordered (failures).

Finally, according to this analogy, comparing the first top 10 features from each vector will correspond to a first draw of 10 balls from the urn, while comparing the top 20 features to a draw of 20 balls, and so on until all balls are drawn at once.

By default the calcHypPI function expects that the top 10% of the features of the two vectors are similarly ordered. This expectation can be modified by the expectedProp argument. When expectedProp is set equal to NULL the number of white balls in the urn (i.e. the top ranking features in the correct order) corresponds to the number of balls that are drawn at each attempt (i.e. the increasing size of top features from each vector that are being compared).

Value

It returns a numeric matrix containing the probability intervals for CAT curves based on equal ranks. The column names of this matrix specifies the quantiles of the hypergeometric distribution used to compute the intervals. The values represent the proportions of overlap associated with the defined quantiles. The resulting matrix object is used to add the probability intervals shades when plotting CAT curves by passing it to the preComputedPI argument of the plotCat function.

Note

This function will take more and more time to run when more and more features are used. For this reason it is convenient to compute the probability intervals separately and store the probability intervals matrix for re-use when plotting the CAT curves.

Author(s)

Luigi Marchionni marchion@jhu.edu

References

Irizarry, R. A.; Warren, D.; Spencer, F.; Kim, I. F.; Biswal, S.; Frank, B. C.; Gabrielson, E.; Garcia, J. G. N.; Geoghegan, J.; Germino, G.; Griffin, C.; Hilmer, S. C.; Hoffman, E.; Jedlicka, A. E.; Kawasaki, E.; Martinez-Murillo, F.; Morsberger, L.; Lee, H.; Petersen, D.; Quackenbush, J.; Scott, A.; Wilson, M.; Yang, Y.; Ye, S. Q. and Yu, W. Multiple-laboratory comparison of microarray platforms. Nat Methods, 2005, 2, 345-350

Ross, A. E.; Marchionni, L.; Vuica-Ross, M.; Cheadle, C.; Fan, J.; Berman, D. M.; and Schaeffer E. M. Gene Expression Pathways of High Grade Localized Prostate Cancer. Prostate, 2011, 71, 1568-1578

Benassi, B.; Flavin, R.; Marchionni, L.; Zanata, S.; Pan, Y.; Chowdhury, D.; Marani, M.; Strano, S.; Muti, P.; and Blandino, G. c-Myc is activated via USP2a-mediated modulation of microRNAs in prostate cancer. Cancer Discovery, 2012, March, 2, 236-247

See Also

See qhyper, plotCat, calcHypPI and computeCat.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
###load data
data(matchBoxExpression)

###the column name for the identifiers
idCol <- "SYMBOL"

###the column name for the ranking statistics
byCol <- "t"

###use lapply to remove redundancy from all data.frames
###default method is "maxORmin"
newMatchBoxExpression <- lapply(matchBoxExpression, filterRedundant, idCol=idCol, byCol=byCol)

###select t-statistics and merge into a new data.frame using SYMBOL
mat <- mergeData(newMatchBoxExpression, idCol=idCol, byCol=byCol)

### compute probability intervals with default values
confInt <- calcHypPI(data=mat)

###structure of confInt
str(confInt)

### compute probability intervals with "expectedProp" set to NULL
confInt2 <- calcHypPI(data=mat, expectedProp=NULL)

###structure of confInt
str(confInt2)

marchion/matchBox documentation built on May 9, 2019, 4:07 p.m.