Morgan Langille
@betascience.bsky.social
Canada Research Chair in Human Microbiomics
at Dalhousie University, Director of the IMR (http://imr.bio), and creator of bioinformatic tools such as PICRUSt.
at Dalhousie University, Director of the IMR (http://imr.bio), and creator of bioinformatic tools such as PICRUSt.
I like this figure to show the point, not perfect correlation, but not random either. academic.oup.com/view-large/f...
PICRUSt isn't meant to predict a single taxa and say it doesn't do anything else. Instead it reports the abundance of genes/pathways that are likely to be there.
PICRUSt isn't meant to predict a single taxa and say it doesn't do anything else. Instead it reports the abundance of genes/pathways that are likely to be there.
academic.oup.com
May 8, 2025 at 6:49 PM
I like this figure to show the point, not perfect correlation, but not random either. academic.oup.com/view-large/f...
PICRUSt isn't meant to predict a single taxa and say it doesn't do anything else. Instead it reports the abundance of genes/pathways that are likely to be there.
PICRUSt isn't meant to predict a single taxa and say it doesn't do anything else. Instead it reports the abundance of genes/pathways that are likely to be there.
The majority of functions where we have a good representation of genomes sequenced do correlate. Lots of examples that don't specifically when looking at an individual strain, but that doesn't take away from the overall predictability. (BTW, picrust is phylogenetic based, not taxonomy).
May 8, 2025 at 4:54 PM
The majority of functions where we have a good representation of genomes sequenced do correlate. Lots of examples that don't specifically when looking at an individual strain, but that doesn't take away from the overall predictability. (BTW, picrust is phylogenetic based, not taxonomy).
I agree that PICRUSt is further from the testable "truth" and that people are likely overconfident in its predictions. It still has its time and place and I try to stress that when I chat with people. If we were all in the same room having a beer/coffee I bet we would agree more than disagree.
May 8, 2025 at 4:24 PM
I agree that PICRUSt is further from the testable "truth" and that people are likely overconfident in its predictions. It still has its time and place and I try to stress that when I chat with people. If we were all in the same room having a beer/coffee I bet we would agree more than disagree.
I think the use of "reliably" is important here. Lots of functions correlate with taxonomy. Taxonomy is useful because people assume function when they see a specific genus or species name. Gene annotation is notoriously poor, but we don't tell people to not do it at all. We caution interpretation.
May 8, 2025 at 2:59 PM
I think the use of "reliably" is important here. Lots of functions correlate with taxonomy. Taxonomy is useful because people assume function when they see a specific genus or species name. Gene annotation is notoriously poor, but we don't tell people to not do it at all. We caution interpretation.
Agreed predicting pathogenicity in E.coli is hard from 16S as is lots of other things that are primarily HGT driven. Doesn't mean PICRUSt can't be used for predicting other functions. Lots of functions correlate with phylogenetics. BTW, what about full-length 16S?
May 6, 2025 at 7:08 PM
Agreed predicting pathogenicity in E.coli is hard from 16S as is lots of other things that are primarily HGT driven. Doesn't mean PICRUSt can't be used for predicting other functions. Lots of functions correlate with phylogenetics. BTW, what about full-length 16S?