Exploring histogram binning for gene ontology data

[noamteyssier]

Opening an issue to see if you have already tested this @artemy-bakulin, but could be interesting to see what the effect of different binning strategies are w.r.t. gene membership binning.

Your current method GeneOntology._build_bin_split is performing an even split over the gene membership, but since that value is more exponentially distributed it could be interesting to instead perform the split using the log2 histogram.

Here's an example of that split with the two methods with a bin size of 5 on GO_BP_2021:

and in log2

[artemy-bakulin]

Yeah, I have tested that with simulated data and split method has higher TNR than hist. The difference is small, though.
I also added a third variant from my initial code and it is even better. It is also much simpler and faster. I will make a pull request of it soon, there I also removed many of your expression preprocessing functions because I could not understand how they work though the task is actually pretty straightforward.