I have shallow shotgun data that I am running Ancom and Ancom BC on. With Ancom i get several differentially abundant species but when i run ANCOM-BC with FDR correction, i lose all of the significant taxa.
Im wondering how to weigh Ancom vs Ancom BC and the taxa i found in ANCOM. I thought that Ancom itself had performed well at correcting for false discovery?
That's a great question! The methods used in ANCOM vs. ANCOM-BC are quite different, but ANCOM-BC has a more robust method for bias correction and controlling FDR by estimating the unknown sampling fractions and correcting the bias induced by their differences among samples.
thank you. something i am wondering then is if i got signifincat values for Ancom but not Ancom BC how much i can trust the positive values from ANCOM?
@kkl45,
This is just my personal opinion. I would trusted ANCOM-BC findings more than ANCOM. ANCOM-BC does some corrections under the hood that could be excluding features that ANCOM finds significant. Hopefully meaning that ANCOM-BC will have less false positives than ANCOM. However, You may still consider features that have really High W scores as significant, if that's important to you. I would warn that sometimes ANCOM reports features with really low W score (Like 1) and if those are not in your ANCOM-BC results, I would say that those features are probably not truly significant.