Errors when using low_res parameter to constrain optimization

[s2hui]

Hello,

Thanks for developing this useful tool!

We currently use Seurat to preprocess our single cell data and I wanted to try using your tool and took the following steps.

1. wrote my seurat object to a loom file
2. read the loom file into python:

adata = sc.read_loom(FILE)

3. loaded the over-clustering assignments @ res=1.2 (computed previously in R/Seurat) into the anndata object:

adata.obs['L1_Round0'] = adata.obs['SCT_snn_res_1.2']

4. updated X_pca to contain the Seurat computed PCA embeddings

adata.obsm['X_pca'] = adata.obsm['pca_cell_embeddings']

5. executed the following (added a lower constraint res = 0.3):

SCCAF_optimize_all(ad=adata, plot=False, min_acc=0.9, prefix ='L1', low_res='SCT_snn_res_0.3', use='pca',n_iter=150)

I get the errors below and I am not sure what is wrong. When I run the above without the low_res parameter it completes 10-12 rounds successfully (however it seems to prefer under clustering, hence wanting to try running it with low_res set).

OUTPUT:

R1norm_cutoff: 0.500000
R2norm_cutoff: 0.050000
Accuracy: 0.000000

Round1 ...
Mean CV accuracy: 0.8799
Accuracy on the training set: 0.9441
Accuracy on the hold-out set: 0.8487
Accuracy on the training set: 0.9506
Accuracy on the hold-out set: 0.8477
Accuracy on the training set: 0.9435
Accuracy on the hold-out set: 0.8421
Max R1mat: 0.652632
Max R2mat: 0.006597
min_acc: 0.842117
Traceback (most recent call last):
File "", line 1, in
File "/usr/local/bin/miniconda2/envs/myPythonEnv/lib/python3.7/site-packages/SCCAF/init.py", line 654, in SCCAF_optimize_all
*args, **kwargs)
File "/usr/local/bin/miniconda2/envs/myPythonEnv/lib/python3.7/site-packages/SCCAF/init.py", line 861, in SCCAF_optimize
zmat = np.minimum.reduce([(R1mat > R1norm_cutoff), conn_mat.values])
ValueError: operands could not be broadcast together with shapes (50,50) (93,93)

Thanks for your help,
shui

[jtuddenham]

Hello,

I am running into similar issues as well. I am wondering if there is any update on this. Alternatively @s2hui did you find a fix for this?

Thanks,
John