search

sankaranlab / SCAVENGE

SCAVENGE is a method to optimize the inference of functional and genetic associations to specific cells at single-cell resolution.
GNU General Public License v3.0
79 stars 34 forks source link

Two analyses of the same code yielded different results #13

Open februaryfang opened 1 year ago

februaryfang commented 1 year ago

Hello,

I run the same code by twice. I did not change anything. And I set the same seed number like your examples. But I got the different result of "true_cell_top_idx" for each cell. Is this a bug? Or is it because of the permutation test method?

Best

fl-yu commented 1 year ago

Hi thank you for using our tool! This is a tricky problem that we are still thinking about how to resolve it properly. We realized that setting a seed is not enough.

Given that network propagation is a stochastic process at each step of the iteration, the probability (information) of each cell can be different even if a stationary state of the entire cell-to-cell network is reached. Although two independent analyses with the same settings result in different true_cell_top_idx, I believe the differences should be very subtle and will not affect the following analysis. Please let me know if have any other questions.