search

AlmaasLab / csdR

Efficient CSD implementation in R
https://almaaslab.github.io/csdR
GNU General Public License v3.0
6 stars 1 forks source link

Quick question #3

Open iorilondon opened 1 year ago

iorilondon commented 1 year ago

Hi, this is a super useful package, so thank you very much. Sorry for the back and forth - answered some of my own questions. I do have one new one: I get the results where you have a higher C score (and low S/D), and those where you have a higher D score (and low S/D) - how would you interpret those results that have both higher C and S scores (but low D scores)?

iorilondon commented 1 year ago

Second question: would you say there is a score level where individual scores becomes significant, or is it just about looking at them relative to each other?

yaccos commented 1 year ago

I would recommend having a look at Figure 1 in the paper by Voigt et al. (https://ntnuopen.ntnu.no/ntnu-xmlui/bitstream/handle/11250/2468176/journal.pcbi.1005739.pdf?sequence=2&isAllowed=y). Here, you can see that the C-,S- and D-scores are antagonistic to each other. A gene pair with a high C-score is highly co-expressed with the same sign under the two conditions. Having a high S-score means that the genes are co-expressed under one condition, but are not co-expressed in the other, which is a quite different situation than with a high C-score.

yaccos commented 1 year ago

The typical analyses for CSD are comparing the scores relative to each other without any notion of statistical significance. The csdR package does not support this directly, but I think I have heard of attempts to compare the C-,S- and D-scores from experimental data to results obtained form randomized data in order to assign statistical significance.

iorilondon commented 1 year ago

Hi,

Thanks so much for the responses. So the majority of the results I have do fit into the framework represented by that figure (high individual pair scores in C, S, or D, with the other two very low - eg C of 4, S and D closer to 0). Does this mean that the ones with higher C and S (and low D - eg C and S both around 4, and D closer to 0) are a little strange, then? From your experience, do you get some results like this (that represent the results that fall into the white space of that figure), or does it imply some kind of issue?

On Fri, 31 Mar 2023, 14:34 Jakob Peder Pettersen, @.***> wrote:

I would recommend having a look at Figure 1 in the paper by Voigt et al. ( https://ntnuopen.ntnu.no/ntnu-xmlui/bitstream/handle/11250/2468176/journal.pcbi.1005739.pdf?sequence=2&isAllowed=y). Here, you can see that the C-,S- and D-scores are antagonistic to each other. A gene pair with a high C-score is highly co-expressed with the same sign under the two conditions. Having a high S-score means that the genes are co-expressed under one condition, but are not co-expressed in the other, which is a quite different situation than with a high C-score.

— Reply to this email directly, view it on GitHub https://github.com/AlmaasLab/csdR/issues/3#issuecomment-1491934841, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZL264MNXQEUQCQRICIMR7TW63MOTANCNFSM6AAAAAAWOUVIAE . You are receiving this because you modified the open/close state.Message ID: @.***>

yaccos commented 1 year ago

It may not necessarily be that strange even though the scores are antagonistic. The C-, S- and D-values are normalized by the standard deviations of the gene expression for that gene pair in order to report a signal-to-noise ratio. For gene pairs where the co-expression standard deviation is low, this will result in higher values of C-, S- and D-values even if the mean co-expression values stay the same.