Optimizing NSF Calculation & Customizing GEMs for large data

[Irrationall]

Hi,

Thank you for developing this great tool! I have been analyzing Visium HD data using Flowsig and have a few questions regarding efficiency and customization.

1. Computational Efficiency of NSF Calculation
I processed my Visium HD data using bin2cell, resulting in approximately 160,000 cells. However, computing the GEMs takes a long time and consumes a significant amount of memory. (The function 'construct_gems_using_nsf' was successfully executed on the example dataset.)

• Are there any parameters or methods to optimize the NSF calculation for better efficiency?
• Would downsampling cells or genes, or adjusting certain parameters in the construct_gems_using_nsf function, improve performance?

2. Customizing TFs or Gene Sets Instead of GEMs
Is there a way to input predefined TFs or gene lists that are important for my research?

• For example, if I want to create a custom "GEM" module that does not follow the exact procedure used in Flowsig, how should I integrate this into the process - any format that I need to follow?
• Is it possible to infer flow network for a single TF or gene?

3. Identifying Key TFs for Inflow & Outflow Regulation
According to your paper, you used Random Forest to determine the top TFs downstream of inflow and Linear Regression to identify the top TFs upstream of outflow. Would it be possible to share the code for this procedure?

4. Mapping Genes to GEMs
I would like to determine which genes are assigned to each GEM. It seems that adata.uns['GEM'] contains this information, but I’m unsure about the exact format. Could you clarify how genes are mapped to GEMs and how to extract them programmatically?

Thanks in advance for your help! I look forward to your guidance.