There are several models that could present a useful comparison to the new approach:
network learned with the true latent confounder present
network learned ignoring the latent confounder.
network learned with an estimated latent confounder using the PC of the data as estimate. Here we need to decide how the number of PC to use or whether to just assume 1
Parsana, Princy, Claire Ruberman, Andrew E. Jaffe, Michael C. Schatz, Alexis Battle, and Jeffrey T. Leek. 2019. “Addressing Confounding Artifacts in Reconstruction of Gene Co-Expression Networks.” Genome Biology 20 (May). https://doi.org/10.1186/s13059-019-1700-9.
network learned on the first iteration of our algorithm. This is a good contrast with the previous one where the only difference is that here we are using residual instead of the actual
Another aspect that may be worth considering is regarding the performance metrics. So far we have been using local measures that compare edges. Other more global metrics may be worth including like Structural Intervention Distance or even with simulations of the intervention.
There are several models that could present a useful comparison to the new approach:
Another aspect that may be worth considering is regarding the performance metrics. So far we have been using local measures that compare edges. Other more global metrics may be worth including like Structural Intervention Distance or even with simulations of the intervention.