Error when running ANCOM analysis

Technical Support
1

Hello,

While running the ANCOM analysis on my 16S data set I get the following error when running

qiime composition ancom --i-table Year1_ANCOM/comp_tax_level6_Year1_Merged_table.qza --m-metadata-file turf_metadata3.tsv --m-metadata-column grouping --o-visualization Year1_ANCOM/comp_tax_level6_Year1_Merged_table.qzv --verbose

/home/dcurtis/miniconda3/envs/qiime2-2019.10/lib/python3.6/site-packages/scipy/stats/sta ts.py:3235: RuntimeWarning: invalid value encountered in double_scalars
f = msb / msw

While qiime composition ancom does produce a result, I am curious as to the reason behind this. While searching the forums I came across this explanation. Does this apply to my data? There are low abundance features and I want to make sure I understand what is going on.

I've attached the .tsv output from comp_tax_level6_Year1_Merged_table.qza which has pseudocounts and the .tsv version without pseudocounts. I've also attached the qiime composition ancom .qzv file for your reference.

Appreciate your help troubleshooting.

-Daniel
feature_table_biom.txt (128.7 KB) feature_table_biom_pseudocounts.txt (128.9 KB) comp_tax_level6_Year1_Merged_table.qzv (557.0 KB)

2

Hi @dancurtis87!

I think so, although I could be mistaken. It is also worth pointing out the distinction between a warning and an error - in this case you are reporting a warning, which is usually less severe than an error. An error usually represents a failure of some kind, while a warning is more like "hey, you should know about this thing."

I think technically this can happen even with high abundance features - its just specifically the case when the abundances are identical across samples. It just so happens that that is going to usually be way more likely for low abundance or rare features, especially after applying the psuedocount.

Check out @mortonjt's suggestion on filtering (in that post you linked to above).

:qiime2:

1 Like
3

Thanks you for the clarification @thermokarst :t_rex: :qiime2: :beers:

1 Like