I tried to differential analysis with ANCOM for many groups (time points) and I get some features significantly different. But I can find where those taxa are belonging? can I infer that those taxa increase or decrease based on the percentile table? Can we use percentile data to make a box plot to describe the trend?
The best place to start with discussions of compositionally is probably the ANCOM paper itself. Keep in mind that the caveat in compositional analysis is that you can’t assume one organism is going up or another is being lost mathematically. (There are some arguments to be made about Occam’s razor and biological inference in the setting of a large number of organisms, but whether its a loss of abundance, a gain, or something else is hard to say.) I tend to look at the volcano plots and percentiles to get a sense of where the organism is relatively more abundant. You could also definitely construct boxplots showing the distribution of the (log) relative abundance of signifiant taxa.
But, since you’re dealing with longitudinal data, you might also consider LMEs. The pro is that they take advantage of hte longitudinal nature of your data which decreases your inter-individual noise. The con is that they don’t address the compositionality. But, food for thought.
@Toan if you want to go the route of LMEs as @jwdebelius suggests, these are available in q2-longitudinal. Another issue with LME is that you need to run a separate model for each organism you want to test; you can use the feature-volatility plugin to identify taxa that change most over time, and use that to select taxa to test with LMEs.