Are q-tune log likelihoods comparable between different inputs?

[lahuuki]

Hi BayesSpace team,
I am working with some Visium data that has proved tricky to cluster. I am trying some different methods for feature selection to select a set of genes to use to calculate PCAs then run clustering with BayesSpace. I have run q-tune and found a clear likelihood peak for either input.

My question is are these likelihood comparable between the input gene sets? For instance the "SVGm" gene set has much higher likelihoods than the "Marker" gene sets for all values of q tested (plot below) , can I infer that SVGm clustering is a better fit for the data?

Thanks!