Count normalization for generate_taxa_trend_test_long()

[AlexanderUm]

I was wondering why options for original LinDa count normalization ("count") as well as possibility to use data as it is ("other") are not options anymore in generate_taxa_trend_test_long()?
Are there specific reasons?

[cafferychen777]

Dear @AlexanderUm,

Thank you for bringing up this important question about the data type options in the generate_taxa_trend_test_long() function.

Background on the Previous Decision

You're right that the original LinDA function supports three data types: "count", "proportion", and "other". However, in our previous implementation of generate_taxa_trend_test_long(), we had limited the options to just "count" and "proportion" for several reasons:

1. Statistical consistency: We wanted to ensure consistent preprocessing across analyses to improve result comparability.
2. User experience: By automatically handling normalization, we aimed to simplify the workflow for users.
3. Error prevention: The CLR transformation used in LinDA requires data without zeros, which can be problematic when using raw data directly.

Recent Update

After careful consideration of your feedback and reviewing the statistical implications, we've now added the "other" option back to the function. This update has been implemented and pushed to the main branch (commit 40970bc).

The updated function includes:

1. Support for all three data types: "count", "proportion", and "other"
2. Automatic zero handling: For the "other" option, we've implemented a half-minimum imputation approach for zeros to prevent CLR transformation issues
3. Clear warning messages: Users selecting the "other" option will receive appropriate guidance about ensuring proper data preprocessing

Usage Notes

When using the "other" option:

• This assumes your data has been appropriately pre-processed
• The function will check for zero values and apply half-minimum imputation if needed
• You'll have more flexibility for custom preprocessing workflows

For most users, we still recommend using the "count" option (which applies TSS normalization) or "proportion" for pre-normalized data, as these ensure proper statistical handling.

Documentation

The function documentation has been updated to include detailed descriptions of all three options:

• "count": Raw count data that will be automatically normalized using TSS
• "proportion": Pre-normalized data (e.g., relative abundance)
• "other": Custom pre-processed data (with appropriate warnings)

Thank you again for your valuable feedback which has helped improve the flexibility of our package while maintaining statistical rigor.

Best regards,
The MicrobiomeStat Team