I have three subsets of a set of genes. Each subset is a ranked list. I want to calculate what the statistically most significant overlap between the three subsets looks like, i.e. how many of the top-ranked genes I need to take from each list.
So far I've thought about using the multivariate hypergeometric distribution (similar to the same problem with two lists), but I couldn't figure yet out how to use it to get a significance score/p-value.
Any ideas how to approach this problem or papers that I should read?
1 answer
It seems like you may be interested in a consensus ranking. Methods like this might help:
http://www.jmlr.org/papers/volume11/meila10a/meila10a.pdf
Unfortunately, I cannot find a software page or github account for the authors where the method might be available.
Log in to answer this question.
-
Testing overlap significance of three gene sets using the hypergeometric distribution
written by Pac314 •
-
Statistical significance of the intersection of three gene lists
written by roy.granit
-
Check the significance for the number of genes in a cluster
written by The Last Word
-
Which DE genes should be included in pathway search
written by jomo018
-
On the significance of the overlap between lists of genes
written by elb
-
Enrichr analysis using ranked gene list
written by halo22
-
Choosing background list in protein set overlap analysis
written by nand •
-
mRNAs targeted by list of miRNAs
written by Megatron •
-
Comparing 2 pathway over-representation analyses
written by weeweedelivery123 •
-
Validity Of Gene List Comparisons For Lists Originating From Different Platforms?
written by Adamc
You're unlikely to want a hypergeometric distribution-based test for ranked lists. I suspect that you want something like a Kolmogorov-Smirnov test. This would then be similar to gene set enrichment, though you'd want to test the opposite (namely for similarity). You might want to ask this on cross-validated.
Could you give an example of your input data and what you expect as output?