scArches reproducing error: shared latent representation lacks ['predictions']

[sukyorsuki]

Dear authors, I recently tried to find a transfer learning method to annotate celltypes in query data. And found an excellent paradigm you published as "A cellular hierarchy framework for understanding heterogeneity and predicting drug response in acute myeloid leukemia". Thank you for sharing such an easy-to-follow approach on Github.

But when I tried to reproduce the"Fig_3.1_scRNA_Relapse_Abbas_et_al.ipynb" after downloading the raw counts of the vanGalen and Abbas cohort from GEO, everything went well unitl the last step in Concatenate session,

full_latent = sc.AnnData(model.get_latent_representation(adata=adata_full))

##ZGC equals to "Abbas"

where I found that the full_latent.obs not containing [predictions], and failed to annotate the Abbas relapsed AML cohort.

I used scArches 0.5.4 , scvi-tools 0.16.6, python 3.9.13, torch 1.10.2

Would you please tell me what's wrong? I would like to share detailed code results if asked. My email adress: kuaiyi-666@163.com
Looking forward to your kind reply!