Neighborhood Graph Construction shared or not

[RunyuXia]

Hi,
I'm using Milo to perform differential abundance testing across different comparisons (e.g., WT vs Condition1, WT vs Condition2). Would you recommend constructing a single shared neighborhood graph for all comparisons, or building the graph separately for each one? I've tried both approaches, using a shared graph versus constructing them individually, and noticed slight differences in the results.