ChIPXpress: ChIPXpress Ranking

In ChIPXpress: ChIPXpress: enhanced transcription factor target gene identification from ChIP-seq and ChIP-chip data using publicly available gene expression profiles

Description

Ranks list of predicted TF bound genes from ChIPx data using a database of publicly available gene expression profiles.

Usage

ChIPXpress(TFID, ChIP, DB, w=0.1, c=0, warn=FALSE, DBmu=NULL, DBvar=NULL)

Arguments

TFID	A character value corresponding to the Entrez GeneID of the TF-of-interest.
ChIP	A ordered vector of Entrez GeneIDs corresponding to the predicted TF bound genes from ChIPx data. Entrez GeneIDs should be reported as character values and sorted from the most likely to the least likely to be bound by the TF based on the strength of peak signals near each gene in ChIPx data.
DB	A big.matrix of expression values that is used as the database of publicly availiable gene expression profiles. This database can be either built by the user directly using the functions 'build database' and 'clean database', or loaded from a pre-built database in package ChIPXpressData. ChIPXpressData currently contains two pre-built databases avaliable for analysis purposes - GPL1261 for mouse ChIPx data and GPL570 for human ChIPx data. See example code below on how to load a big.matrix formatted gene expression database.
w	A numeric value specifying the weight used to combine the ChIPx and gene expression compendium based rankings. The typical user will not need to modify the default value. Specifically, if we let Pg be the rank of gene g based on ChIPx data and Ag be the rank of gene g based on gene expression compendium data, the final ChIPXpress score for gene g is w*Pg+(1-w)*Ag. ChIPXpress rankings are then sorted from the smallest to the largest baed on the ChIPXpress scores. Thus, for w < 0.5, the gene expression information has a larger impact on the final ChIPXpress rankings, and for w > 0.5, the input ChIPx rankings has a larger inpact on the final ChIPXpress rankings. Tests using real datasets revealed that w=0.1 is the optimal weight.
c	A numeric value specifying the TF expression cutoff. The typical user will not need to modify the default value. The TF expression cutoff, c, is used to remove samples in which the TF expression is below the TF expression cutoff, c, prior to the calculation of the absolute correlation. Testing using real datasets revealed that c=0 is the optimal TF expression cutoff.
warn	If set to TRUE, then will report warning if mean, variance, and coefficient of variation of the TF expression.
DBmu	Mean of each probe in the database prior to standardization. Can be found for the pre-built databases in the ChIPXpressData package.
DBvar	Variance of each probe in the database prior to standardization. Can be found for the pre-built databases in the ChIPXpressData package.

Details

ChIPXpress works by ranking the TF bound genes predicted from ChIPx data by the absolute correlation between each gene and the TF in the database of gene expression profiles. Genes that are more highly correlated, either positively or negatively, are ranked as the most likely to be an actual TF target gene. Note, the absolute correlation is calculated only using the samples in which the TF expression is above c=0, i.e. the mean TF expression across all samples, in order to improve prediction performance. See reference below for more information on the rationale behind the ranking algorithm of ChIPXpress and the selection of the default values w=0.1 and c=0.

Value

Returns a list with two vectors: the first vector contains the absolute correlations of each predicted TF bound gene with the TF in ranked order, where the names of the vector correspond to the Entrez GeneID of each gene, and the second vector contains the Entrez GeneIDs of the predicted TF bound genes not found in the database. An additional warning will be given if DBvar is specified, which will let the user know if the TF has a low expression variance in the database.

Author(s)

George Wu

References

Wu G. and Ji H. (2012) ChIPXpress: enhanced ChIP-seq and ChIP-chip target gene identification using publicly available gene expression data. In preparation.

Examples

## Example analyses of real Oct4 bound genes predicted
## from ChIP-seq data in ESC using pre-built GPL1261 
## database

## Load predicted Oct4-bound genes from ChIPx data
data(Oct4ESC_ChIPgenes)

## Load example GPL1261 Database
library(bigmemory)
path <- system.file("extdata",package="ChIPXpressData")
DB_GPL1261 <- attach.big.matrix("DB_GPL1261.bigmemory.desc",path=path)
## To load the human GPL570 data, replace 'DB_GPL1261' with 'DB_GPL570'.

## Run ChIPXpress ("18999" is Entrez GeneID of Oct4)
out <- ChIPXpress(TFID="18999",ChIP=Oct4ESC_ChIPgenes$EntrezID,DB=DB_GPL1261)
head(out[[1]]) ## ChIPXpress Rankings
head(out[[2]]) ## Missing genes not found

## For the final step, you can convert the Output into a 
## clean table with genes names or any other preferred gene identifier 
## by using any of your favorite annotation packages (e.g., biomaRt). 
## Here, we can use the original Oct4ESC_ChIPgenes dataframe to do so directly.
GeneNames <- Oct4ESC_ChIPgenes$Annotation[match(names(out[[1]]),Oct4ESC_ChIPgenes$EntrezID)]
Result <- data.frame(1:length(out[[1]]),GeneNames,names(out[[1]]),out[[1]])
colnames(Result) <- c("Rank","GeneNames","EntrezID","ChIPXpressScore")
head(Result) ## Clean ChIPXpress rankings

ChIPXpress documentation built on Nov. 8, 2020, 4:58 p.m.