Sparsity_sc and sparsity_sp = 0

[gboscagli]

Greetings,

I'm using Tangram to map a scRNA-seq reference of mouse lung into a spatial dataset (Visium 10X) of the same tissue. Both datasets are preprocessed using standard Scanpy pipelines. I followed the Tangram tutorial available in the Squidpy docs, so I segmented the cells using watershed and used the resulting cell count to tailor the target_count and density_prior for map_cells_to_space, as shown in the linked tutorial, with mode="constrained". The training genes are 1K, as expected. Once I visualize the diagnosis plots using plot_training_scores, however, the result is the following.

Basically, all the training scores are between 0.952 and 0.999 because sparsity_sc and sparsity_sp are both 0 for all genes, and the predictions on test genes yield an auc_score of 0. This issue happens for other two pairs of scRNA-seq/spatial datasets that I'm using, so I think I'm missing some steps in the pipeline.

I'm wondering which could be the reason for such behaviour. Hope you can help me, I would really appreciate it!