search

MarioniLab / milo

Differential abundance on KNN graphs
GNU General Public License v3.0
1 stars 0 forks source link

How robust is Milo to different topologies? #11

Open MikeDMorgan opened 4 years ago

MikeDMorgan commented 4 years ago

This is related to need (or not?) for batch correction prior to graph construction. Related to issue #3 - how different do the topologies need to be for Milo to break down/fail?

emdann commented 3 years ago

I think this is quite tough to quantify and it will still be challenging to account for ALL possible topologies that can be found in different real datasets.

MikeDMorgan commented 3 years ago

I guess the original motivation here was really about how much a batch or technical effect impacted on DA testing, but also what if the perturbation or condition just leads to such a different manifold that cells really don't overlap? The situation is hypothetical and deliberately pathological - I don't know how often you would get 2 different manifolds for the same cells but from different experimental, biological or medical contexts.

Maybe park this for now - but it might be a question reviewers also pose.