Set individual exponent with .json or .tsv.

[ACastanza]

The current behavior is to use a constant as the exponent for weighting the K-S or AUC calculation. Under certain circumstances it may be desirable to set this exponent on a per-gene basis.

To do so, it might be useful to allow this parameter to alternatively accept a json file containing a key-value pair where each gene(key) is assigned the exponent(value) that should be used when weighting that gene.

Ideally the current behavior to allow a constant for standard usage would be preserved as well.

[KwatMDPhD]

exponent = number | .tsv of the same shape as the feature_x_sample_x_(numbner|metric).tsv, where corresponding feature_x_sample_x_exponent.tsv applies to feature_x_sample_x_(...).tsv.

[KwatMDPhD]

| .json feature-to-exponent (this is useful as you said for people who have a saved weighting system.)

[ACastanza]

exponent = number | .tsv of the same shape as the feature_x_sample_x_(numbner|metric).tsv, where corresponding feature_x_sample_x_exponent.tsv applies to feature_x_sample_x_(...).tsv.

In most cases I think we would only need one possible exponent for each gene, so it would be more the shape of the gene_x_metric_score matrix, so just a two column file where the first column is the gene and the second column is the weighting exponent to use for that gene.

I suppose this might be different for ssGSEA where maybe you'd want the ability to set a different exponent for different samples, but I think if that were to be the case the samples could probably be run separately.

[ACastanza]

| .json feature-to-exponent (this is useful as you said for people who have a saved weighting system.)

I don't actually have a preference for using TSV or json for this, I'd just suggested json because you seemed to like using that structure, and in python it has some benefits in its easy conversion to dict

[KwatMDPhD]

This is more involved than expected. Do you think this is something people can do in pre-rank fashion? They can make their vector is whatever way they wish?

[ACastanza]

I don't think manually applying it to preranked data would really do what I'd want it to do

[KwatMDPhD]

Got it. Will make it happen. Excited to see your use case!