TypeError when using the exporting.spring_project function

[tejas-j]

I've been trying to export my data to SPRING using the external export feature. I do all my processing in Seurat and then export a loom file to be read in scanpy and export to SPRING. I've been however getting the below error.

The notebook with the code and the loom file can be found here: https://github.com/tejas-j/seurat2spring

Is there a better way to achieve what I'm trying to do (data -> Seurat -> scanpy -> SPRING)

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-13-3b0044b18ade> in <module>
      1 import time
      2 t0 = time.time()
----> 3 sce.exporting.spring_project(data, './pbmc3k', 'draw_graph', subplot_name='force1', overwrite=True)
      4 print(time.time() - t0)

~/miniconda2/envs/py36/lib/python3.6/site-packages/scanpy/_exporting.py in spring_project(adata, project_dir, embedding_method, subplot_name, cell_groupings, custom_color_tracks, total_counts_key, overwrite)
    157     # Write continuous colors
    158     continuous_extras['Uniform'] = np.zeros(E.shape[0])
--> 159     write_color_tracks(continuous_extras, subplot_dir / 'color_data_gene_sets.csv')
    160 
    161     # Create and write a dictionary of color profiles to be used by the visualizer

~/miniconda2/envs/py36/lib/python3.6/site-packages/scanpy/_exporting.py in write_color_tracks(ctracks, fname)
    301     out = []
    302     for name,score in ctracks.items():
--> 303         line = name + ',' + ','.join(['%.3f' %x for x in score])
    304         out += [line]
    305     out = sorted(out,key=lambda x: x.split(',')[0])

~/miniconda2/envs/py36/lib/python3.6/site-packages/scanpy/_exporting.py in <listcomp>(.0)
    301     out = []
    302     for name,score in ctracks.items():
--> 303         line = name + ',' + ','.join(['%.3f' %x for x in score])
    304         out += [line]
    305     out = sorted(out,key=lambda x: x.split(',')[0])

TypeError: must be real number, not numpy.str_

[flying-sheep]

Hi @swolock, any ideas?

[tejas-j]

@calebweinreb any ideas on how to fix this issue?

[davidhbrann]

I think the spring export function currently fails because it only checks whether each column in adata.obs is a pandas categorical variable (not is_categorical(adata.obs[obs_name])) and, if not, assumes it's a continuous variable and then tries to join a str with an integer.

If you look at your file data.obs contains a number of categorical variables that are currently numpy objects

data.obs.dtypes
ClusterID             int32
ClusterName          object
RNA_snn_res_0_5      object
nCount_RNA          float32
nFeature_RNA          int32
orig_ident           object
percent_mt          float32
seurat_clusters      object
louvain            category
dtype: object

As a quick fix, I think you can do something like this:

adata = data.copy()
obj_cols = adata.obs.columns[adata.obs.dtypes == np.object]
adata.obs[obj_cols] = adata.obs[obj_cols].astype('category') 
sce.exporting.spring_project(adata, './pbmc3k', 'draw_graph', subplot_name='force1', overwrite=True)

Not sure what's the best way to fix it for the future: check for other dtypes or uses f-strings to avoid the str concatenation errors?

[LuckyMD]

doesn't sanitize_anndata() do exactly that @flying-sheep? Could just add a call to that before exporting to SPRING?