Suggestion: Standardize `model.predict()` output format across all models

[tnyamaguchi]

Hi team,

Thanks for the great work on biolearn. The modular model interface and GeoData integration make it really convenient to work with methylation-based predictors.

While using multiple models in a batch prediction workflow, I noticed that the model.predict() output format varies depending on the model:

1. Most models return a DataFrame where:

• Sample IDs are stored as the row index (pandas .index)
• The predicted value is stored in a single "Predicted" column

2. Some models (e.g. Zhang_10) return sample IDs as a regular "id" or "SampleID" column, rather than as the DataFrame index

3. Models like TwelveCellDeconvoluteBloodEPIC return transposed matrices:

• Rows = cell types
• Columns = sample IDs (requiring a .T to match other models)

To ensure downstream code (e.g. batch pipelines, pd.concat, or tidy reshaping) works reliably, would it be possible to standardize the .predict() output for all models?

Really appreciate all the effort you've put into this framework!

[sarudak]

Glad you're enjoying the library. Thanks for pointing out this inconsistency. All the deconvolution models have this issue and we will change them to be consistent with the other models.

I was not able to replicate the Zhang_10 behavior you described, it looks just like the other model outputs when I run it.

[sarudak]

To fix all deconvolution clocks should be transposed and their expected test results should be transposed also. This should affect TwelveCellDeconvoluteBloodEPIC, DeconvoluteBloodEPIC and DeconvoluteBlood450K