getGeneSim: Compute functional similarity for genes

Description

Calculate the pairwise functional similarities for a list of genes using different strategies.

Usage

getGeneSim(genelist1, genelist2=NULL, similarity="funSimMax", similarityTerm="relevance", normalization=TRUE, method="sqrt", avg=(similarity=="OA"), verbose=FALSE)

Arguments

genelist1

character vector of primary gene IDs according to organism annotation package (see setEvidenceLevel)

genelist2

optional other character vector of primary gene IDs to compare against

similarity

method to calculate the functional similarity between gene products

similarityTerm

method to compute the similarity of GO terms

normalization

normalize similarities yes/no

method

"sqrt": normalize sim(x,y) <- sim(x,y)/sqrt(sim(x,x)*sim(y,y)); "Lin": normalize sim(x,y) <- 2*sim(x,y)/(sim(x,x) + sim(y,y)); "Tanimoto": normalize sim(x,y) <- sim(x,y)/(sim(x,x) + sim(y,y) - sim(x,y)). NOTE: normalization does not have any effect, if term similarity is NOT "relevance" and similarity = "funSimMax", "funSimAvg" or similarity = "OA" and avg=TRUE

avg

standardize the OA kernel by the maximum number of GO terms for both genes

verbose

print out some information

Value

n x n similarity matrix (n = number of genes)

Details

The method to calculate the pairwise functional similarity between gene products can either be:

"max": the maximum similarity between any two GO terms
"mean": the average similarity between any two GO terms
funSimMax: the average of best matching GO term similarities. Take the maximum of the scores achieved by assignments of GO terms from gene 1 to gene 2 and vice versa. [2]
funSimAvg: the average of best matching GO term similarities. Take the average of the scores achieved by assignments of GO terms from gene 1 to gene 2 and vice versa. [2]
"OA": the optimal assignment (maximally weighted bipartite matching) of GO terms associated to the gene having fewer annotation to the GO terms of the other gene. [1]
"hausdorff": Hausdorff distance between two sets: Let X and Y be the two sets of GO terms associated to two genes. Then $dist(X,Y) = \max\{\sup_{t \in X} \inf_{t' \in Y} d(t,t'), \sup_{t' \in Y} \inf_{t \in X} d(t,t') $ [3]. Since GOSim 1.2.8 we translate the Haussdorff distance into a similarity measure by taking $sim(X,Y) = \exp(-dist(X,Y)$.
"dot": the dot product between feature vectors describing the absence/presence of each GO term. The absence of each GO term is weighted by its information content. Depending on the type of later normalization one can arrive at the cosine similarity (method="sqrt") or at the Tanimoto coefficient (method="Tanimoto").[4]

References

[1] H. Froehlich, N. Speer, C. Spieth, A. Zell, Kernel Based Functional Gene Grouping, Proc. Int. Joint Conf. on Neural Networks (IJCNN), 6886 - 6891, 2006.

[2] A. Schlicker, F. Domingues, J. Rahnenfuehrer, T. Lengauer, A new measure for functional similarity of gene products based on Gene Ontology, BMC Bioinformatics, 7, 302, 2006.

[3] A. del Pozo, F. Pazos, A. Valencia, Defining functional distances over Gene Ontology, BMC Bioinformatics, 9:50, 2008.

[4] M. Mistry, P Pavlidis, Gene Ontology term overlap as a measure of gene functional similarity, BMC Bioinformatics, 9:327, 2008.

Examples

Run this code

	# see evaluateClustering

Run the code above in your browser using DataLab