`compute_covet()`: log1p-transformed data incorrectly detected as raw counts, causing double log-transformation

[NoeCoursi]

Hello!

Thank you for developing ENVI and this new promising approach to spatial transcriptomics.

While integrating COVET into a pipeline, I came across what looks like an edge case in compute_covet() and wanted to flag it.

In line 400 of utils.py, the heuristic used to detect whether spatial_data.X is already log-transformed relies on checking whether the minimum value is negative:

if spatial_data.X.min() < 0:
    # Data is already log-transformed
    print("Using expression data from X (appears to be log-transformed)")
    ...
else:
    print("Log-transforming expression data from X")
    exp_data = np.log(spatial_data[:, CovGenes].X + 1)

As far as I understand, this transformation should not produce values below 0. However, I believe that using a log1p transformation is quite standard in single-cell and spatial transcriptomics (e.g., sc.pp.log1p()). Notably, the function itself creates log1p-transformed layers (lines 364 and 406) using np.log(X + 1).

Although this is not a critical bug, and the print statement does indicate which branch is taken, I feel like the message "Log-transforming expression data from X" might not raise suspicion if the user assumes the detection is reliable.

I'm not sure how much this affects COVET outputs, but I wanted to bring it to your attention.

Maybe an integer detection like np.all(np.mod(X, 1) == 0) or an additional parameter could help avoid this ambiguity.

Hope this help !