How To Evaluate The A Newly Developed Clustering Algorithm?

Question

[ftp]

Hello,

I developed a new clustering algorithm specific for gene expression for gene function prediction... I'm interested in assessing the validity of my method through some biological datasets. Is there a dataset where it lists genes which have similar functions? I want to check if genes which have similar functions are clustered together....

thanks

[David Westergaard]

If you are clustering on gene expression levels alone, I fail to see why two genes with similar function would cluster. Just because two genes are kinases, they most likely won't have the same expression levels. At least I've never seen evidence for such.

Also, how do you define "similar function"? Do two genes have similar function if they're both kinases? If they're part of the same pathway? If they're both transmembrane? etcetcetc. If you define similar function by pathway, there's plenty of data sets.

[ftp]

Can you direct me to one of the pathway datasets? and yes i define similar function by pathway

[David Westergaard]

MSigDB is probably the most easy to parse. There are pathways from both KEGG and REACTOME.

Answer 1

[Istvan Albert]

The way to evaluate the algorithm is to generate random data, cluster it then verify that the data that you have in the clusters do indeed belong there.

Trying to evaluate the quality of clustering from a second, possibly unrelated attribute is not the correct approach.

[ftp]

Ok I understand that and I will definitely do that....However, its also important to check the validity of the method when applied to gene expression as the goal is to have a set of clusters where genes belonging to the same cluster share similar function....Is there a dataset that gives me genes that have known functions?