Aggregating SCENIC results without Nextflow

[igrisch]

Hi all,

Due to some funkiness with the cluster I'm working on, I can only use the Docker implementation of pySCENIC but I can't use Nextflow. I ran the whole SCENIC process just fine, and I completed the process 10 times; however, now I'm not sure about best-practices for aggregating the runs. I have never used Nextflow and have a hard time visually parsing what's happening in those scripts. So all the threads that just point me back to the Nextflow pipeline are not super helpful. Could someone explain in plain terms what's happening so I can reverse-engineer it? Or is it complex enough that I should just give up on doing things manually?

My current thought process is:

1. Run the grn step 10-100x
2. Read in the adjacency matrix results to Python or R and tally the number of times a TF-target pair occurs. Keep those pairs that appear in >80% of runs
3. Unsure about this one: For those that I keep, do I average the results of "importance" column in the adjacency matrix?
4. Write out a custom adjacency matrix file using the pairs that have passed muster.
5. Run the rest of the pipeline using this custom adjacency matrix file.

Any ideas, or reasons this approach doesn't work?