We recently analysed a reference standard of known composition using MetaPhlAn3 and found that the relative abundance values for many of the taxa deviated quite a bit from the ground truth. We believe this is a MetaPhlAn3 problem as the same data analysed by a commercial provider using a different tool yielded abundance values much closer to the ground truth. On the flip side, MetaPhlan3’s false positive and false negative rates were excellent.
Which MetaPhlAn3 parameters can / should we tweak to improve the software’s abundance estimates? I understand any tweaking may have detrimental effects on false positive and false negative rates.