Error in run_ulm: No Sources with More than min_n=2 Targets Despite Matrix-Network Compatibility

[victorsanchezarevalo]

Describe the bug
I am encountering an error when using Decoupler's run_ulm function. The error states that no sources have more than min_n=2 targets, even though I have reduced the min_n parameter, and my dataset should have sufficient shared targets between the matrix (mat) and the network (collectri).

Error:

File ~/miniforge3/envs/decoupler/lib/python3.10/site-packages/spyder_kernels/customize/utils.py:209 in exec_encapsulate_locals
  exec_fun(compile(code_ast, filename, "exec"), globals)

File ~/Documentos/Mis_analisis/Ester_Martin/12_decoupler.py:225
  tf_acts, tf_pvals = dc.run_ulm(mat=mat, net=collectri, verbose=True, min_n=2)

File ~/miniforge3/envs/decoupler/lib/python3.10/site-packages/decoupler/method_ulm.py:108 in run_ulm
  net = filt_min_n(c, net, min_n=min_n)

File ~/miniforge3/envs/decoupler/lib/python3.10/site-packages/decoupler/pre.py:146 in filt_min_n
  raise ValueError("""No sources with more than min_n={0} targets. Make sure mat and net have shared target features or
ValueError: No sources with more than min_n=2 targets. Make sure mat and net have shared target features or
  reduce the number assigned to min_n

To Reproduce
Steps to reproduce the behavior:

1. Install decoupler in a clean Python environment.
2. Use the run_ulm function with the following setup:
   • Input matrix (mat): Gene expression matrix with n_genes x n_samples.
   • Regulatory network (collectri): A list of transcription factors and their target genes.
3. Set min_n=2 to ensure that there are at least 2 targets per transcription factor.
4. Run the analysis and observe the error when filt_min_n fails to find sufficient shared targets between the matrix and network.

If needed, I can provide a subset of the data that triggers the error for testing purposes.

Expected behavior
I expected the run_ulm function to return transcription factor activities and p-values when using the provided gene expression matrix (mat) and regulatory network (collectri), as there should be sufficient shared target genes between the two.

System

• OS: Fedora 40
• Python version: 3.10

Additional context
I have verified that my matrix contains valid gene names and is compatible with the regulatory network. Despite lowering min_n to 1, the issue persists. This error seems to indicate a mismatch between the features in the matrix and the network, but these have been checked for consistency.

[PauBadiaM]

Hi @victorsanchezarevalo,
decoupler follows the observations x features convention (commonly used in Python), rather than the features x observations convention (more typical in R). You can transpose your matrix to match this format, and it should work. Feel free to reach out if you have any further questions!

[victorsanchezarevalo]

Hi,

I’m encountering an issue when running ULM analysis with decoupler. I have already transposed my expression matrix as recommended (observations x features), but I am still getting the following error related to the min_n=2 parameter:

Code used:

# Preparing matrix and transposing
mat = results_df[['stat']].T.rename(index={'stat': 'treatment.vs.control'}).T

# Transposing matrix so genes are in rows
print(f"New mat shape: {mat.shape}")

# Assigning gene names from subset_adata.var['gene_name']
mat.index = subset_adata.var['gene_name'].values

# Checking the first 10 gene names
print(mat.index[:10])

# Retrieving CollecTRI gene regulatory network
settings.setup(curl_timeout=1200)
os.system('rm -rf ~/.cache/omnipathdb/*')
os.system('rm -rf ~/.cache/pypath/*')

collectri = dc.get_collectri(organism='mouse', split_complexes=False)

# Running ULM analysis with min_n=2
tf_acts, tf_pvals = dc.run_ulm(mat=mat, net=collectri, verbose=True, min_n=2)

# Checking results
print(tf_acts.head())
print(tf_pvals.head())

Error message:

ValueError: No sources with more than min_n=2 targets. Make sure mat and net have shared target features or reduce the number assigned to min_n

I have verified that the matrix has been transposed and gene names have been correctly assigned, but the issue persists. It seems that no transcription factors have more than 2 shared targets, even though min_n=2 is a reasonable threshold in this context.

Any help or suggestions would be appreciated!

Thank you!

[PauBadiaM]

Could you show me the head of your input mat?

mat.head()

[victorsanchezarevalo]

mat.head()

Out[20]: 
         treatment.vs.control
Sox17               -0.252628
Gm15452              0.320179
Gm26983              0.747064
Gm6187              -0.106062
Gm6119              -0.746242

[PauBadiaM]

Hi @victorsanchezarevalo,

As you show in your console output you have one observation (one contrast) and multiple genes (n). So, your matrix has wrong format features x observations (n, 1), not the correct observations x features (1, n). Transpose it again and it should be fine.