I am getting the zero balance error when my balances aren't zero...
QIIME version: qiime2-2017.10
qiime gneiss ols-regression \
--p-formula "Species+Soil_new" \
--i-table balances.qza \
--i-tree hierarchy.qza \
--m-metadata-file my-metadata.tsv \
Error: Detected zero variance balances...
Balance table summary: no zeros (attached)
feature-table_summary.txt (7.8 KB)
I explored both feature table (prior to gneiss) and the balanced feature table produced via the ilr-transform command but cannot locate a feature only 0. Any help would be appreciated.
It could be possible that there are zero variance balances. Could you upload your balances or your
Thanks for your help. Attached is the composition table.
comp_root_n10_nmc_uMnU.qza (1.8 MB)
@balford looks like your table has a ton of singletons and doubletons, which is known to be problematic whenever you run compositional methods (including ANCOM). Could you try to filter out features that have less than 10 counts in the total experiment? If an ASV only appears 1 time, its likely garbage.
Thanks for the input. I have re filtered the data for N10 (zero balance error) and now N20 (zero balance error- comp table attached). Due to the large number of samples, should I go up to N100? I have a few hundred samples, should I filter for features which only appear in 5 or more samples?
How is ols-regression calling a "zero balance error"? Is it possible that there is something else inherent my data that is triggering the error. If so, let me know and we can have an offline discussion.
comp_root_N20_nmc_nMnU.qza (1.4 MB)
I think that is reasonable - if a feature only present in 5 samples across the entire experiment, you really can’t learn much about that feature to begin with. There is just not enough information. I’m looking at your data, and it looks like there are quite a few features that are only present in 1 sample. When filtering out features that only appear in 5 samples or fewer, the number of features half. To me, this suggests that there is some contamination (but would require additional investigation).
Right now, that error is being thrown whenever the variance is zero.
Basically, this means that there are some features that are not actually adding additional information into the analysis, and will cause issues when trying run the regression. So features that aren’t actually present, or only present once in the datasets can cause these sorts of issues.
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.