# question about balance_taxonomy output from gneiss

Hi all,

I’m learning and using gneiss to reveal the taxonomic differences between treatment and control samples, but I could not understand balance_taxonomy output clearly.
For instance, it says “The taxa in the y0numerator on average have increased between patient group and the healthy control.” Does it mean the numerators increased in both groups or just in one group (patient or control)?
I just got the balance_taxonomy out and the numerator/denominator list, but do not know which group (treatment samples or control samples) they belong.
Could you explain for me please?

Xiaolan

Hi @Xiaolan_Lin, feel free to post your output and we can help interpret.

The balance is a ratio - so it doesn’t yield information about who increased / decreased.
Rather, it shows how the ratio has changed between groups, and what microbes make up that ratio.

This should be thought more like concentration, where you don’t ask if either water or salt (for example) is changing, but instead if the ratio of salt / water is changing.

Hi @mortonjt,

Thank you for your explanation! I can understand the balance as a ratio now, but I still have no idea whether treatment or control was regarded as numerator?
Here I post my y2_taxa_summary result. y2_taxa_summary.qzv (142.5 KB) To my understanding, the numerator is control so the ratio would be control/BaP. The taxa in y2numerator increased more with control samples (proportionally). Am I get it right?
I also post the regression result here B-regression_summary.qzv (4.1 MB) and another question is, am I right to choose y2 to do the following balance-taxonomy summary? And the R squared value is quite small that I am worried if gneiss is appropriate for my data.

Hi @Xiaolan_Lin, you’re getting warmer concerning the interpretation. The balance is a ratio with regards to the microbes, but not the samples - you should try to avoid inferring if microbes in the numerator / denominator are changing, but rather if the ratio as a whole is changing across samples.

In short, your R2 is not great - you should be aiming for >10% in this sort of context.
I’d checkout Aldex2 or Songbird - you may find the output more readable.

Hi @mortonjt,