how to evaluate my biomedical literature mining method?

Question

[amira_h]

Hello All,

I'm new to bioinformatics and I've been working on a biomedical literature mining method to predict human genes relation.
I have produced my data and would like now to compare it to other methods to test the performance of my systems (Recall, Precision, etc) I find it difficult to choose these methods, could you recommend online methods that I can use (preferably if they provide good API documentation)

I'm currently trying to compare my method with the methods included in GOSemSim Package

, thank you

[Lars Juhl Jensen]

It depends entirely on which type or types of associations you are trying to predict. You need to ensure that you are comparing methods that actually try to do the same thing, or it is bound to become an unfair/meaningless comparison.

Answer 1

[Jean-Karim Heriche]

I found iHOP to be reasonably good at linking human genes with similar biological function. I think it is worth comparing to. Contact the authors to get bulk data. You can also download text mining data from STRING but I found it to be much less good at linking human genes with similar function. I would also suggest to compare to prediction using protein-protein interactions because that is known to contain most information for human genes or to some state of the art prediction approach like GeneMANIA.