saezlab/liana-py

Question regarding ranking and aggregating from all methods.

Marwansha opened this issue · 5 comments

Marwansha commented

I have a general question and sorry if its a naive one

i have 2 different PBMC samples (disease 1 and disease 2) and i did independently for each sample li.mt.rank_aggregate function, and filtered based on specificity_rank for significance (>0.01)

can we interpret a ligand-receptor pair interaction , that is significant in 1 sample not the other , having a high magnitude_rank in the significant condition and low in the other , can we say that this ligand -receptor pair interaction is condition / sample specific ?

take as an example in png below the scatter plot of magnitude_rank values from each condition, the pair of ligand -receptor interaction highlighted in circle on top left, it is significant ( >0.01 specificity_rank) only in this cov samples , with high magnitude scores, and low magnitude score in the other condition with non significant_specificity_rank ) .

can i say this ligand -receptor interaction is condition specific ?

thanks

image

dbdimitrov commented

Hi @Marwansha,

I guess, just to state the obvious, there is no way to (statistically) quantify differences between two conditions with a single sample per condition.

But yes, I assume you could say that according to the rankings (which can be interpreted as a probability of an interaction being highly ranked in a given sample/disease), there is an indication of some interactions being specific to one of your samples.

If this is of any value, a similar and simple approach was also used by a colleague of mine in the past (in a very similar scenario).

Hope this helps! :)

Marwansha commented

Thanks a lot good to know that ver similar scenario was already used by a colleague and actually it's not 1 sample per condition i just grouped all the different samples into their corresponding conditions " around 130 sample per condition"

so i hope i can move on with my approach , thanks again

dbdimitrov commented

Hi @Marwansha,

Ah in this scenario, I should mention, that we also provide 3 different ways to compare interactions across conditions.

You may consider using either one of those also: 😉

  • 👍1
Marwansha commented

Thanks alot @dbdimitrov this is very helpful will give them a try.

Have an amazing day

  • ❤️1
dbdimitrov commented

I will close the issue, feel free to open new ones :)