Error: The usage_norm is NaN

[BioinfoXP]

Everthing I do before these following steps is ok, howerver I can not solve the error

The usage_norm is NaN, I can not do with the downstream analysis.

In addition, this step also make an error

[weiju455]

Check whether the matrix used for cnmf has rows and columns that are all zeros

[MDMotlagh]

I had this issue very recently as well. The consensus usage.k_ .txt file is returned with cell names in rows and k number of components in columns with empty matrix and that is why we see NaN for usage_norm values. I think it is relate to the incompatibility with newer versions of numpy.

I tackled this by creating a conda env and installing the following package versions:
pip install scanpy==1.9.4 pandas==1.5.1 fastcluster==1.2.5 cnmf==1.3 PyYaml==6.0 numpy==1.24

conda list:
_libgcc_mutex 0.1 main
_openmp_mutex 5.1 1_gnu
anndata 0.11.4 pypi_0 pypi
array-api-compat 1.12.0 pypi_0 pypi
bzip2 1.0.8 h5eee18b_6
ca-certificates 2025.2.25 h06a4308_0
cnmf 1.3.0 pypi_0 pypi
contourpy 1.3.2 pypi_0 pypi
cycler 0.12.1 pypi_0 pypi
exceptiongroup 1.3.0 pypi_0 pypi
expat 2.7.1 h6a678d5_0
fastcluster 1.2.5 pypi_0 pypi
fonttools 4.58.5 pypi_0 pypi
h5py 3.14.0 pypi_0 pypi
joblib 1.5.1 pypi_0 pypi
kiwisolver 1.4.8 pypi_0 pypi
ld_impl_linux-64 2.40 h12ee557_0
libffi 3.4.4 h6a678d5_1
libgcc-ng 11.2.0 h1234567_1
libgomp 11.2.0 h1234567_1
libstdcxx-ng 11.2.0 h1234567_1
libuuid 1.41.5 h5eee18b_0
libxcb 1.17.0 h9b100fa_0
llvmlite 0.44.0 pypi_0 pypi
matplotlib 3.10.3 pypi_0 pypi
natsort 8.4.0 pypi_0 pypi
ncurses 6.4 h6a678d5_0
networkx 3.4.2 pypi_0 pypi
numba 0.61.2 pypi_0 pypi
numpy 1.24.0 pypi_0 pypi
openssl 3.0.16 h5eee18b_0
packaging 25.0 pypi_0 pypi
pandas 1.5.1 pypi_0 pypi
patsy 1.0.1 pypi_0 pypi
pillow 11.3.0 pypi_0 pypi
pip 25.1 pyhc872135_2
pthread-stubs 0.3 h0ce48e5_1
pynndescent 0.5.13 pypi_0 pypi
pyparsing 3.2.3 pypi_0 pypi
python 3.10.18 h1a3bd86_0
python-dateutil 2.9.0.post0 pypi_0 pypi
pytz 2025.2 pypi_0 pypi
pyyaml 6.0 pypi_0 pypi
readline 8.2 h5eee18b_0
scanpy 1.9.4 pypi_0 pypi
scikit-learn 1.7.0 pypi_0 pypi
scipy 1.15.3 pypi_0 pypi
seaborn 0.13.2 pypi_0 pypi
session-info 1.0.1 pypi_0 pypi
setuptools 78.1.1 py310h06a4308_0
six 1.17.0 pypi_0 pypi
sqlite 3.45.3 h5eee18b_0
statsmodels 0.14.4 pypi_0 pypi
stdlib-list 0.11.1 pypi_0 pypi
threadpoolctl 3.6.0 pypi_0 pypi
tk 8.6.14 h993c535_1
tqdm 4.67.1 pypi_0 pypi
typing-extensions 4.14.1 pypi_0 pypi
tzdata 2025.2 pypi_0 pypi
umap-learn 0.5.9.post2 pypi_0 pypi
wheel 0.45.1 py310h06a4308_0
xorg-libx11 1.8.12 h9b100fa_1
xorg-libxau 1.0.12 h9b100fa_0
xorg-libxdmcp 1.1.5 h9b100fa_0
xorg-xorgproto 2024.1 h5eee18b_1
xz 5.6.4 h5eee18b_1
zlib 1.2.13 h5eee18b_1

[dylkot]

Thanks for letting me know about this. I will look into why this is happening. Does anyone know what version of numpy creates this error? And if it is reproduced on the example notebook?

[somsomikim]

Hi, I am also struggling with the same problem. By the way, in my case, I have set cnmf.prepare parameter for # of hvg = # of input genes, since I have already selected hvgs to run cnmf. Is there anyone having the same case with me?

[dylkot]

if you want you can email me the input matrix you are using that is giving these errors and the exact sequence of commands and I can try to reproduce. dylkot@gmail.com

[somsomikim]

@dylkot Hi, thanks for your reply. I guess I have found the reason why usage values are all NAs. In my case, I have used selected genes (batch-aware hvgs) for cnmf and set # of hvgs = # of selected genes, however, in refit-usage step, I assume standard deviation of some genes that I have selected were 0 so that usage values got NA. I can use refit-usage=FALSE, but I will try after removing genes with std=0 and let me try new attempt. I will share the results as soon as I got them. I am not sure whether other cases are due to this. Thx!

[dylkot]

ahh thanks for this! I can add a warning to that effect in the next version

[somsomikim]

@dylkot Hi, I have changed selected genes for std > 0, and it solved NAs with the usage files. :) I appreciate your effort on the package.