ubic.erminej.data

Class Multifunctionality

java.lang.Object

ubic.erminej.data.Multifunctionality

public class Multifunctionality
extends Object

Implementation of multifunctionality computations as described in Gillis and Pavlidis (2011) PLoS ONE 6:2:e17258.

Note: This computes multifunctionality at 'startup' based on all the annotations available, not necessarily the ones selected for analysis. For example, if you load KEGG groups along with GO, multifunctionality is computed based on both of them. But if the user selects only to analyze KEGG, multifunctionality correction is based on the bias in KEGG+GO. This could be fixed by changing the 'subclone' method of GeneAnnotations to trim aspects, not just genes.

Version:

$Id: Multifunctionality.java,v 1.38 2013/05/20 02:21:16 paul Exp $

Author:

paul

See Also:

for another implementation of parts of this.

Constructor Summary

Constructors

Constructor and Description
Multifunctionality(GeneAnnotations go, StatusViewer m) Construct Multifunctionality information based on the state of the GO annotations -- this accounts only for the probes in the annotations.

Method Summary

Methods

Modifier and Type	Method and Description
Map<Gene,Double>	adjustScores(Map<Gene,Double> geneToScoreMap, boolean useRanks, boolean weight) Given a set of scores for genes, adjust them to correct for multifunctionality, using a regression approach.
double	correlationWithGeneMultifunctionality(List<Gene> rankedGenes)
double	correlationWithGoTermMultifunctionality(List<GeneSetTerm> rankedGoTerms)
double	enrichmentForMultifunctionality(Collection<Gene> genesInSet) This is like correlationWithGeneMultifunctionality(List), but without an order list to start with, and intended for cases where the number of genes is smaller than the total number of genes.
double	enrichmentForMultifunctionalityPvalue(Collection<Gene> genesInSet)
QuantileBin1D	getGeneMultifunctionalityQuantiles() Get QuantileBin1D, which can tell you the quantile for a given value, or the expected value for a given quantile.
double	getGOTermMultifunctionality(GeneSetTerm goId)
double	getGOTermMultifunctionalityPvalue(GeneSetTerm goId)
double	getGOTermMultifunctionalityRank(GeneSetTerm goId)
Gene	getMostMultifunctional(Collection<Gene> genes) Convenience method.
double	getMultifunctionalityRank(Gene gene)
double	getMultifunctionalityScore(Gene gene)
int	getNumGenes() How many genes have multifunctionality scores.
int	getNumGoTerms(Gene gene)
Double	getRawGeneMultifunctionalityRank(Gene gene)
boolean	isStale()
Map<Gene,Double>	padogWeights()
void	setStale(boolean stale)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

Multifunctionality

public Multifunctionality(GeneAnnotations go,
                  StatusViewer m)

Construct Multifunctionality information based on the state of the GO annotations -- this accounts only for the probes in the annotations. Genes with no GO terms are completely ignored.

Parameters:

go - These annotations should already be pruned down to those used in analysis.

Method Detail

padogWeights

public Map<Gene,Double> padogWeights()

Returns:

weights computed on multifunctionality as per Tarca et al. (2012)

adjustScores

public Map<Gene,Double> adjustScores(Map<Gene,Double> geneToScoreMap,
                            boolean useRanks,
                            boolean weight)

Given a set of scores for genes, adjust them to correct for multifunctionality, using a regression approach. If the regression coefficient (slope) is negative, no correction will be done.

Parameters:

geneToScoreMap - Should already be log transformed, if requested.

useRanks - If true, the ranks of the gene scores will be used for regression.

weight - If true, the regression will be weighted (current implementation is by 1/sqrt(rank))

Returns:

residuals from the regression, which are to be used as the new scores.

correlationWithGeneMultifunctionality

public double correlationWithGeneMultifunctionality(List<Gene> rankedGenes)

Parameters:

rankedGenes, - with the "best" gene first.

Returns:

the rank correlation of the given list with the ranks of the multifunctionality of the genes. A positive correlation means the given list is "multifunctionality-biased". Genes lacking GO terms are ignored.

correlationWithGoTermMultifunctionality

public double correlationWithGoTermMultifunctionality(List<GeneSetTerm> rankedGoTerms)

Parameters:

rankedGoTerms, - with the "best" GO term first.

Returns:

the rank correlation of the given list with the ranks of the GO term multifunctionality of the terms. A positive correlation means the given list of terms is "multifunctionality-biased".

enrichmentForMultifunctionality

public double enrichmentForMultifunctionality(Collection<Gene> genesInSet)

This is like correlationWithGeneMultifunctionality(List), but without an order list to start with, and intended for cases where the number of genes is smaller than the total number of genes. We do this the same we we do for GO groups. Implementation of algorithm for computing AUC, described in Section 1 of the supplement to Gillis and Pavlidis; see http://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U.

Parameters:

genesInSet -

Returns:

the ROC AUC for the genes in the set compared to the multifunctionality ranking

See Also:

in baseCode for a generic implementation

enrichmentForMultifunctionalityPvalue

public double enrichmentForMultifunctionalityPvalue(Collection<Gene> genesInSet)

Parameters:

genesInSet -

Returns:

The pvalue associated with the ROC AUC for the genes in the set compared to the multifunctionality ranking, based on the Mann-Whitney U test / Wilcoxon

getGeneMultifunctionalityQuantiles

public QuantileBin1D getGeneMultifunctionalityQuantiles()

Get QuantileBin1D, which can tell you the quantile for a given value, or the expected value for a given quantile.

Returns:

getGOTermMultifunctionality

public double getGOTermMultifunctionality(GeneSetTerm goId)

Parameters:

goId -

Returns:

the computed multifunctionality score for the GO term.

This is computed from the area under the ROC curve for the genes in the group, in the ranking of all genes for multifunctionality. The exact method for computing this is defined by computeGoTermMultifunctionalityRanks

getGOTermMultifunctionalityPvalue

public double getGOTermMultifunctionalityPvalue(GeneSetTerm goId)

Parameters:

goId -

Returns:

the pvalue associated with the multifunctionality of the gene set

See Also:

getGOTermMultifunctionality

getGOTermMultifunctionalityRank

public double getGOTermMultifunctionalityRank(GeneSetTerm goId)

Parameters:

goId -

Returns:

the relative rank of the GO group in multifunctionality, where 1 is the highest multifunctionality, 0 is lowest. WARNING, this does not correct for the presence of multiple GO groups with the same genes (redundancy)

getMostMultifunctional

public Gene getMostMultifunctional(Collection<Gene> genes)

Convenience method.

Parameters:

genes -

Returns:

the gene with the highest multifunctionality

getMultifunctionalityRank

public double getMultifunctionalityRank(Gene gene)

Parameters:

gene -

Returns:

relative rank of the gene in multifunctionality where 1 is the highest multifunctionality, 0 is lowest

getMultifunctionalityScore

public double getMultifunctionalityScore(Gene gene)

Parameters:

gene -

Returns:

multifunctionality score. Note that this score by itself is not all that useful; use the rank instead; or for a "human-readable" version use the number of GO terms, which this approximates (in terms of ranks). Higher values indicate higher multifunctionality

getNumGenes

public int getNumGenes()

How many genes have multifunctionality scores.

Returns:

getNumGoTerms

public int getNumGoTerms(Gene gene)

Parameters:

gene -

Returns:

number of GO terms for the given gene.

getRawGeneMultifunctionalityRank

public Double getRawGeneMultifunctionalityRank(Gene gene)

Parameters:

gene -

Returns:

rank. Where zero is the highest rank, ties accounted for.

isStale

public boolean isStale()

setStale

public void setStale(boolean stale)