Include targets with negative correlation to TFs?

[rcw1]

I noticed that each regulon contains only targets that are positively correlated to the regulon's TF. Many TFs, including those in which I am interested, are repressors so it would be very valuable to keep the negatively correlated targets. In runSCENIC_2_createRegulons.R, line 69 keeps only targets with positive correlation to the TF: tfModules_Selected <- tfModules_asDF[which(tfModules_asDF$corr==1),]

Is there a reason that you chose to include only positively-correlated targets? Is there a parameter I can change to include both positively and negatively-correlated targets? Or do I need to alter the code?

This seems like a great package, but this issue is a real sticking point for me. Thanks for your help.

Best,

Rob

[eegk]

Hi Rob, I'm also using scenic and there is a strong bias towards positive correlations. I'm using also negative correlations and the distribution of TFs - Targets is almost entirety positive, do you observe the same phenomena?. I wonder if the SCENIC team can further detail in this?

Thank you

[vkartha]

Hi - anyone have a response to this? I notice that there is a Spearman filter of 0.03 in the runSCENIC_1_coexNetwork2modules function. I saw that the tutorial says :

GENIE3/GRNBoost can detect both positive and negative associations. In order to distinguish potential activation from repression, we will split the targets into positive- and negative-correlated targets (i.e. Spearman correlation between the TF and the potential target).

in the runCorrelation step, but I only realized after that it filtered out negative associations. (I was following the steps listed under Command List here

Is there a way to leave negative associations in? Would it make a difference to the way SCENIC estimates TF-gene associations (given what is being discussed above)?