Hi all,

I fitted a MEFISTO (MOFA2) model with three views (three omics layers, methylation, rna, and protein) and a shared continuous time covariate. Two factors clearly capture time - both decrease over time across all groups (bioreactors = biological replicates) (and I also annotated the factors with the help of a linear model and the meta data, and both are significant for the time meta info).

When I run GSEA on the factor loadings, the same GO term shows positive NES for Factor 1 for the protein omics layer view and negative NES for Factor 2 for the protein omics layer view. If both factors go down with time, this seems impossible: a pathway can't be both up- and down-regulated over time.

Has anyone seen opposite NES for the same term in temporal factors, and how did you handle factor orientation for consistent interpretation?

Thanks!