In "Supplementary Note 22: Hyperparameter Search" there's the following part:
"[...] the number of top similar genes in the co-expression network H_pert – {3, 5, 10, 20}; the number of top similar genes in the perturbation network H_gene – {3, 5, 10, 20}; [...]"
Is there a mix-up here? According to the main text, the co-expression network is used for deriving the gene embedding, while the GO-term network is used to derive the perturbation representation. What is the number of of neighbors used in the co-expression and GO-term networks for the final model?
In "Supplementary Note 22: Hyperparameter Search" there's the following part:
"[...] the number of top similar genes in the co-expression network H_pert – {3, 5, 10, 20}; the number of top similar genes in the perturbation network H_gene – {3, 5, 10, 20}; [...]"
Is there a mix-up here? According to the main text, the co-expression network is used for deriving the gene embedding, while the GO-term network is used to derive the perturbation representation. What is the number of of neighbors used in the co-expression and GO-term networks for the final model?