Cluster identification/exploration with RSS and AUC scores

[hbandukw]

Hi,

I am currently using SCENIC to identify some clusters and explore others in my data. I was wondering how to ask the right questions which could allow me to have a more directed exploration.

1. Biological interpretation of the z-score clustered heatmap

If (i) the clustered heatmap is based on the z-scores of the AUC matrix. AUC measures the activity of the regulon (no. of TG in top 5% of ranked cell/ no. of target genes in regulon); and (ii) cells containing similar regulons cluster together; then

Meaning:

• these regulons have similar target genes and are most likely connected in a network?
• these regulons can have different target genes but are all most associated with these clusters of cells?
• summary statistics of the regulon distribution is used to cluster the cells in the clustered heatmap?

2. RSS scores I was wondering if you can use Jensen–Shannon divergence to figure out the association of the given regulons (2 at a time)?

3- Using the binarized auc_mtx (with auc_thresholds) Let's say that I want to ask the following questions: what regulons are highly expressed or present in cluster-X in dataset-S ?

In the binarized auc_mtx, can you claim the following about the regulon that has the highest no. of 1's in that group: basically say something like "In cluster X , regulon Y has the highest no. of cells that are marked as active (relative to regulon Y) compared to regulon Z which has the second highest (again, compared to itself)?

Thanks a lot!