Also: reading the methods is usually a good way to understand what's happening. This is exactly the case

This is how it's done

academic.oup.com/bioinformati...

Oh, yes, I see your point.

Well, thanks for the material.
Also, why are you skeptical?

I must admit a couple of things:
1- I have no idea what that means (except for "single-cell multimodal data")
2- I struggle in keeping the pace with the methods behind modern omics sciences
3- I have to find time and resources for some math classes

They are everywhere

I guess that if the attention window is large enough, the ranking becomes locally less important, so it's relevant for a gene to be in the top/middle/bottom part of the list.
Is it just a supersized GSEA(ish), then?

I understand that many models are based on ranking genes by their expression, anything that affects ranking may affect the performance, or not?

Diagonal: cat and wolf are animals I like, dog and lion not so much

He’s not the first one who seems to forget that, at a certain point, an experiment has to be performed.
Also, biology is not all about single cells and perturbations

Well, gene models also depend on genome build, GENCODE for GRCh37 is frozen at what version?
And then there’s selection on the non coding part… yes, WTF?

www.bbc.com/news/article...

How?