allez: Random-set calibration of gene-set statistics

In atbroman/allez: Random-set calibration of gene-set statistics

Description

For each set (i.e. category) in a collection defined by Gene Ontology (GO) or the Kyoto encyclopedia of genes and genomes (KEGG), allez computes a standardized score that measures how unusual microarray measurements in that set are compared to measurements in same-sized random subsets of the microarray-level data. allez may be used to assess the enrichment of a category for genes that are interesting, for example, owing to differential expression across groups or owing to substantial correlation with some other phenotype. Additionally allez may be used to identify sets that show unusual variance characteristics.

Usage

allez(scores, lib, idtype = c("ENTREZID", "SYMBOL"), library.loc = NULL,
sets = c("GO", "KEGG"), locallist = NULL,
collapse = c("full","partial","none"),
reduce = NULL, setstat = c("mean", "var"),
universe = c("global", "local"),
transform = c("none", "binary", "rank","nscore"),
cutoff = NULL, annotate = TRUE)

Arguments

scores	numeric vector of microarray-level or organism-level scores, usually measuring differential expression among groups or the relationship of expression values to some other variable: possibly log fold change, t-statistic, indicator of significant differential expression (i.e. gene list), posterior probability of differential expression, or correlation with some phenotype. For microarray-level scores, the vector must be named with manufacturer probe names; for example, scores must be named by the Affymetrix probe set IDs. For organism-level scores, the vector must be named with Entrez Gene Identifiers.
lib	character string, name of data package corresponding to microarray (e.g. "hgu133plus2"), or organism ("org.Hs.eg"). A vector of package names is allowed in case multiple chips are used (e.g "moe430a" and "moe430b" cover the mouse genome.)
idtype	character string, either "ENTREZID" (default) or "SYMBOL", corresponding to the type of names used for the scores vector.
library.loc	a character vector describling the location of R library trees to search through, or 'NULL'. The default value of 'NULL' corresponds to all libraries currently known to '.libPaths()'. Non-existent library trees are silently ignored.
sets	character string, describing the collection of sets. "GO" (default) or "KEGG"
locallist	list contains in-house gene sets of interest. Default is NULL. Each element in the list represents a gene set. Each element should be a character string containing genes' entrez IDs (if idtype="ENTREZID") or gene symbols (if idtype="SYMBOL"). Element names will be used as set names. If locallist is not NULL, universe="local" will be disabled.
collapse	character string, describing the method for reducing probe-level data to gene-level data. "full" (default) uses the function reduce to reduce from probe level to gene level. This uses the ENTREZID class associated with the data package/s. "none" probe level results "partial" adjustment to the z-scores computed in "full" by the factor sqrt n.genes/n.probes .
reduce	function to collapse/summarize gene-level scores; if reduce=NULL and scores are binary 0,1, the default is max, if scores are continuous the default is median. reduce=mean is also appropriate for continuous scores.
setstat	"mean" or "var" indicating the function used to compute unstandardized set scores from microarray level scores.
universe	If "global", each set's score is compared to the score of a random set taken from all annotated genes (if collapse="full" or scores are organism-level) or probes (otherwise). If "local", comparison is made not to the entire set of annotated probes/genes but to those defining different parents in the GO DAG (so it's not applicable with KEGG). With multiple parents per set, the one with the largest Z score is reported.
transform	optional transformation of microarray-level data. If "binary", a cutoff must also be supplied. Transformation is done after collapsing to gene level if collapse="full". "nscore" is the normal scores transformation qnorm( rank( ./(G+1) ) ) , which makes the gene set means very close to normal, and thus improves the z-score quality
cutoff	numeric cutoff when transform = "binary"; selected genes are those with probe/gene score larger than (or equal to) cutoff
annotate	logical, whether to include set names in the output

Details

allez uses formulas for both the expected value and variance of a sample mean or sample variance computed on a random subset of fixed microarray-level data. These formulas enable it to standardize observed scores computed on categories from GO or KEGG, and thus to make the different categories comparable in terms of how unusual they are compared to random sets. Facilities allow various microarray-level scores, various set-level scoring methods (mean, variance), various reductions to gene-level, and various calibrations (i.e. for GO, should we use the whole annotated collection of genes/probes or just those in a GO parent).

Value

returns a list, containing components:

setscores

data frame with rows for gene sets and columns for summary information scoring these sets. Information includes set name (if annotate = TRUE), the set sample mean ( or set sample variance if setstat="var" ), the set size (number of probes if collapse="none" and number of genes otherwise), and the appropriate Z-score (nominally distributed as a standard normal variate). If a functional set contains the entire set of inputted gene scores, the z-score will output NA.

aux

a list, with auxiliary information from the calculation, including a data frame set.data, which contains, in table format, the map of each gene to a gene set and the gene score. The globe vector is the full complement of annotated probes (or genes) (this is the ‘universe’ if universe="global").

If universe="local" and sets="GO", aux also contains a matrix recording set-level results for all parents of every child set (the setscores in this case reports only the largest Z-score among all the parents).

Author(s)

Michael Newton, Deepayan Sarkar, Aimee Teo Broman, Subhrangshu Nandi

References

Newton, M.A., Quintana, F.A., den Boon, J.A., Sengupta, S., and Ahlquist, P. (2007). Random-set methods identify distinct aspects of the enrichment signal in gene-set analysis. Annals of Applied Statistics, 1, 85-106.

Sengupta, S., den Boon, J.A., Chen, I.H., Newton, M.A. et al. (2006). Genome-wide expression profiling reveals EBV-associated inhibition of MHC class I expression in nasopharyngeal carcinoma. Cancer Research, 66, 7999-8006.

Examples

data(npc)
scores  <- (1/2)*sqrt(28)*log((1-npc)/(1+npc))

npc.kegg <- allez( scores=scores, lib="hgu133plus2", sets="KEGG")
allezTable(npc.kegg)[,1:4]

atbroman/allez documentation built on May 10, 2019, 2:08 p.m.