I have analyzed my data with ancom and tried to compare the output from --p-transform-function 'log', 'sqrt', and 'log'.
However, only the 'clr' (default) transform function resulted in a graphical element in the visualization; the other two functions resulted in visualizations, where the graph resulted in the error "Uncaught SyntaxError: Unexpected token I in JSON at position 34584", while the tables below seemed to be ok.
Is this error related to ANCOM or to the visualization of the qzv?
taxa-collapsed-L5-pseudocount-log.qzv (2.0 MB) taxa-collapsed-L5-pseudocount-sqrt.qzv (2.0 MB) taxa-collapsed-L5-pseudocount-clr.qzv (2.0 MB)
Hey @arwqiime, looks like the
log results both contain
Infinity values in them, which is causing this to break. What isn't clear to me though is how we are winding up with
Infinity, since in theory the imputation step should handle that (I think?). Would you be willing to share your
FeatureTable[Frequency] (QZA with UUID
48cf99c5-5256-428f-9d29-faf0bd8f5370) file with us? You can send a link in a DM to me if you aren't comfortable sharing publicly. Thanks!
Hi @thermokarst, of course, here are the tables:
The first one is a taxa-collapsed table (the one you requested), the second one is this one with a pseudocount added to it.
taxa-collapsed-L5.qza (1.6 MB) taxa-collapsed-L5-pseudocount.qza (1.6 MB)
Thank you for your support!
Hi @thermokarst, did you have a chance to look at the requested data for the ancom visualisation issue (sqrt, log).
Hi @arwqiime - @ebolyen has taken over this thread for me, he spent some time looking at this yesterday, and will update you once he has some ideas or questions. Thanks!
May I add another question to this ANCOM thread. We have analysed 16S amplicons in laboratory scale bioreactors for ca. 2 month; samples were taken at day, 17, 32, and 50, (day after satellite start; DASS) respectively, from 5 independent reactors.
We observed that the microbial community started to segregate over time, and the communities at day 50 were considerably different.
In order to identify the most differing taxa at the various time points, we first filtered the samples by the time points and did independent ancom analysis only for the samples at the various time points.
W considerably increased from day 17 to 50, as well as clr increased, because the differences in the community were greated at day 50, which was already visible in beta diversity analyses.
However, at day 50, only one single taxon was listed in the ancom statistical results section next to the ANCOM volcano plot, while more taxa were flagged at day 32 and 17 as true (reject null hypothesis). This does not make sense to me. If W and clr increase both, then I would expect more taxa listed in the ancom statistics table, and not less.
I have added the three visualizations below. ancom_Ferm-Type_L7_DASS-17_addpc.qzv (449.8 KB) ancom_Ferm-Type_L7_DASS-32_addpc.qzv (429.7 KB) ancom_Ferm-Type_L7_DASS-50_addpc.qzv (428.4 KB)
Hi there @arwqiime, really sorry that this fell off of our radar.
I have submitted a fix, hopefully it will be included in the 2020.5 release of QIIME 2:
The reason that you are seeing this error is because the
transform-function can potentially produce “divide by zero” results in certain cases (for example, when you impute a zero count by 1, and then take the log of that, which is