pseudotimespace_global() always yields "Connectivity is undefined for the null graph" exception

[marcouderzo]

Hello,
I have a strange issue with running st.spatial.trajectory.pseudotimespace_global() on one particular dataset, which is a spatial subset of a full Visium HD sample. This is my code:

adata_gd75 = ad.read_h5ad("analysis_output/adata_subset.h5ad")
adata_gd75 = st.convert_scanpy(adata_gd75)

adata_gd75.obs['clusters'].value_counts()

clusters
7     490
16    383
6     325
14    271
22    269
0     260
4     249
15    241
10    226
2     224
21    215
8     212
5     176
1     169
17    153
20    111
12    100
9      82
13     74
23     63
27     48
3      42
29     38
28     35
11     28
19     23
25     18
30      7
24      7
32      7
18      5


sc.pp.pca(adata_gd75, n_comps=30, random_state=30)
sc.pp.neighbors(adata_gd75, n_neighbors=15, use_rep="X_pca")
sc.tl.umap(adata_gd75, random_state=30)
sc.pl.umap(adata_gd75, color="clusters")

adata_gd75.uns["iroot"] = st.spatial.trajectory.set_root(adata_gd75,use_label="clusters",cluster=21,use_raw=False)
st.spatial.trajectory.pseudotime(adata_gd75,eps=50,use_rep="X_pca",use_label="clusters")    # runs indefinitely

At first, the pseudotime() call never finished running, and kept one of my CPU cores at 100% for three days straight. By analyzing the call stack, it kept recursing around those nx calls:

networkx/algorithms/simple_paths.py: all_simple_paths(...)
networkx/algorithms/simple_paths.py: all_simple_edge_paths(...)
networkx/algorithms/simple_paths.py: _all_simple_edge_paths(...)

The same code, on a subset of another sample, worked just fine and finished the computation in less than an hour. I then found out that by taking a wider subset of the adata_gd75 subset, the call finished running and returned some trajectories. Maybe it wasn't finding any trajectories at all?

However, afterwards, I run:

for lin in lineage:
    lin = [str(c) for c in lin]
    lineage_str = array_to_string(lin)
    try:
        st.spatial.trajectory.pseudotimespace_global(adata_gd75,use_label="clusters", list_clusters=lin)
        st.pl.cluster_plot(adata_gd75,use_label="clusters", 
                           list_clusters=lin, 
                           show_trajectories=True,
                           show_subcluster=True, 
                           figsize = [12,12], 
                           cmap='cet_glasbey',
                           fname = f"./figures/all_samples/lineage_tracing/GD7.5_niches_subset/GD75_{lineage_str}_pseudotimespace_global.png")
    except Exception as e:
        print(f'{lineage_str}:{e}')

This yields:

...

Start to construct the trajectory: 28 -> 17 -> 15
28-17-15:Connectivity is undefined for the null graph.
Start to construct the trajectory: 28 -> 16 -> 1
28-1-16:Connectivity is undefined for the null graph.
Start to construct the trajectory: 2 -> 28 -> 16
28-2-16:Connectivity is undefined for the null graph.
Start to construct the trajectory: 28 -> 4 -> 16
28-4-16:Connectivity is undefined for the null graph.
Start to construct the trajectory: 28 -> 16 -> 7
28-7-16:too many indices for array: array is 1-dimensional, but 2 were indexed
Start to construct the trajectory: 28 -> 12 -> 16
28-12-16:Connectivity is undefined for the null graph.
Start to construct the trajectory: 28 -> 16 -> 15
28-15-16:Connectivity is undefined for the null graph.

...

So I'm going to ask: in stLearn, what causes it to yield this "Connectivity is undefined for the null graph." exception?

Any advice? Thanks.

Best,
Marco Uderzo