Differing shapes in sctransform_from_parameters

[Normann-BPh]

Hello,

I've encountered a confusing error. As in, I don't think this should happen if the input has been processed thus far.

I run ComSeg via a script-version of your tutorial, with SRT Xenium data and Xenium cell masks.
After image-processing I receive ValueError: operands could not be broadcast together with shapes (93187,541) (93187,539) in sctransform_from_parameters, as you can see below.

Is this error caused by a wrong-package version or actually mismatching input?

The only other option I could imagine is, that something is changing self.anndata between passing fit_params, array_of_vect to sctransform_from_params and calculating param_sctransform from the count_matrix, but that sounds too convoluted to be the problem.

Hope you can help!

Kindest regards
n-bph

(ComSeg) [user]$ python scripts/seg_comseg.py
config dict overwritting default parameters
11693
prior added to transcripts_36_pixel_relative and saved in csv file
dict_centroid added for transcripts_36_pixel_relative
Processing images: 100%|████████████████████████████████████████| 1/1 [06:15<00:00, 375.49s/it]
100%|████████████████████████████████████████████████████████| 541/541 [02:01<00:00,  4.45it/s]
  0%|                                                                    | 0/1 [00:00<?, ?it/s]improvement of modularity 0.006270115694725686
improvement of modularity 0.00036438836270757946
improvement of modularity 2.210827103554802e-05
improvement of modularity 2.3542855431823995e-07
stop because of improvement of modularity 5.517261314480493e-09  below  the threshold  1e-07
100%|██████████████████████████████████████████████████████████| 1/1 [22:26<00:00, 1346.10s/it]
Writing temporary files...
Running scTransform via Rscript...

Attaching package: arrow

The following objects are masked from package:feather:

    read_feather, write_feather

The following object is masked from package:utils:

    timestamp

Calculating cell attributes from input UMI matrix: log_umi
Variance stabilizing transformation of count matrix of size 539 by 10000
Model formula is y ~ log_umi
Get Negative Binomial regression parameters per gene
Using 539 genes, 10000 cells
  |======================================================================| 100%
Found 12 outliers - those will be ignored in fitting/regularization step

Second step: Get residuals using fitted parameters for 539 genes
  |======================================================================| 100%
Calculating gene attributes
Wall clock passed: Time difference of 16.19295 secs
Reading output files...
Clipping residuals...
user/micromamba/envs/ComSeg/lib/python3.8/site-packages/comseg/utils/preprocessing.py:59: RuntimeWarning: divide by zero encountered in log10
  regressor_data[:, 1] = np.log10(np.sum(array_of_vect, axis=1))
Traceback (most recent call last):
  File "scripts/seg_comseg.py", line 69, in <module>
    Comsegdict.run_all(
  File "user/micromamba/envs/ComSeg/lib/python3.8/site-packages/comseg/dictionary.py", line 580, in run_all
    self.compute_insitu_clustering(
  File "user/micromamba/envs/ComSeg/lib/python3.8/site-packages/comseg/dictionary.py", line 258, in compute_insitu_clustering
    self.global_anndata = self.in_situ_clustering.cluster_rna_community(
  File "user/micromamba/envs/ComSeg/lib/python3.8/site-packages/comseg/clustering.py", line 138, in cluster_rna_community
    count_matrix_anndata = sctransform_from_parameters(self.param_sctransform,
  File "user/micromamba/envs/ComSeg/lib/python3.8/site-packages/comseg/utils/preprocessing.py", line 63, in sctransform_from_parameters
    res = (array_of_vect - mu) / np.sqrt(mu + mu ** 2 / fit_params[0, :])
ValueError: operands could not be broadcast together with shapes (93187,541) (93187,539)