carmonalab / UCell

Gene set scoring for single-cell data
GNU General Public License v3.0
135 stars 16 forks source link

Is there a way to see how much each gene contributes to the UCell score? #6

Closed ksaunders73 closed 3 years ago

ksaunders73 commented 3 years ago

Hello!

As the title says, is there a way to see in a quantifiable way how much each gene contributes to the UCell score? Or is it assumed that both contribute equally or that the score is a sort of average of the two, or the absolute expression is altered such that higher scored genes are weighted over the other?

For example, say gene x has a "true" score of 0.5 and gene y has a "true" score of 1.0. When brought together into the UCell score, and their individual contributions are calculated after the UCell score's calculation, will their individual contributions to the score be 0.5 and 1.0 respectively, or both be 0.75 (the average of the individual "true" scores), or something where gene y will be weighted more because it has a higher UCell score? Does that make sense?

Thanks for reading!

mass-a commented 3 years ago

That's a good question. There are a few caveats (e.g. how draws in ranking are dealt with), but in practice it would be a good approximation to consider the UCell score of a signature as the average of UCell scores for individual genes. The formula for U scores is fairly simple (see the paper), so you can try plugging in some numbers yourself.

To answer your question, it would be fair to estimate how much each gene in a signature contributes to a UCell score by calculating UCell scores for individual genes.

ksaunders73 commented 3 years ago

That makes sense, thank you very much for your help!