Any suggestion for combining different types of scores

Question

[Na Sed]

I am investigating the similarity between gene list A as reference and another gene list, called B.

I measure the similarity between them in different aspects, such as their GO similarity, DO similarity, the number of literature which have detected them in a specific disease, etc.

I'd like to find a way to integrate all of such scores together such that I combine all of scores in one quantity.

Do you have any suggestion about the combining methods? or have you seen such papers? I appreciate if you could help me.

[Istvan Albert]

I would advise against collapsing different quantities into a single score - as attractive as it may sound the pitfalls and potential to generate misleading results is just as high.

[Jean-Karim Heriche]

Indeed you can't just take any set of scores and combine them and expect to still get a valid measure of similarity. For example, if one of the scoring function is not symmetric, i.e. S(a,b)!=S(b,a) then a combination including it is not guaranteed to be interpretable as a similarity measure. However, this being said, data integration in this way using kernels has already a long history in bioinformatics.

Answer 1

[Jean-Karim Heriche]

If your similarity measures are valid kernel functions then the (weighted) average is still a similarity measure (because it's still a kernel, some other combinations also produce valid kernels). For an example, see my paper here.