Is it seasonable to not filter genes by qval_threshold 0.05 in fs.pp.determine_informative_variables) but remove insignificant genes after fs.tl.learn_intercellular_flows()？

[RuiyingChenBioinfo]

Hi flowsig team,

I'm a fan of your fantastic tool. We firstly tried flowsig in our data with standard parameters in all functions, and we noticed that, with some insignificant genes removed in fs.pp.determine_informative_variables (with qval_threshold set to 0.05 as default), some GEMs will not be linked to any inflow or outflow genes, and are therefore removed from the learned graph. As these GEMs are quite important in our previous results, that is not something we don't want to happen.

Here I'm asking if it is appropriate to not remove these genes, or use less stringent q-value threshold (such as 0.25, as it is sometimes used in other published articles) in fs.pp.determine_informative_variables(), and remove the insignificant genes from the finally generated flow plots. This change will keep more GEMs for further analysis.

Best wishes,
Ruiying

[axelalmet]

Hi Ruiying,

Thank you for your interest in FlowSig! To answer your question, I think it is perfectly fine to change the q-value threshold when removing informative genes. You can also try changing the logFC threshold to be lower, if that would help.

If you are interested in keeping certain GEMs, you can also directly apply fs.pp.filter_flow_vars(filter_flow_vars, vars_to_subet), where vars_to_subset is a list that includes the list of GEMs you want to retain for inference.

Hope that helps! Let me know if you have any other questions.

Best wishes,
Axel.

[RuiyingChenBioinfo]

Hi Axel,

Thank you for your very timely answer! I feel my question is answered well.

Best wishes,
Ruiying