I am trying to use the Differential Abundance Analysis to interpret my data. When I try to run the qiime gneiss ols-regression command, I encounter the error:
Plugin error from gneiss:
cannot convert float NaN to integer
Debug info has been saved to /tmp/qiime2-q2cli-err-8lbj_2kc.log
My commands were:
qiime gneiss ols-regression --p-formula “subject+duration+efficiency” --i-table qiime2-balances/balances.qza --i-tree qiime2-sleep-balances/hierarchy.qza --m-metadata-file metadata.tsv --o-visualization qiime2-balances/regression_summary.qzv
Does it mean I have a problem with my metadata?
Yes, please see this topic. You are probably missing values from the metadata categories you are using.
I hope that helps!
I took a look at the link. It seems that my metadata is ok. I validated it for qiime2 with Keemei.
Could it be something else?
Keemei will not necessarily catch this… this issue with empty metadata values is a problem specific to gneiss (because the OLS method cannot handle missing values)
However, if you have checked the file yourself and confirm that there are not missing values, then please read through that entire thread that I linked to. It is a long one, but there is a lot of debugging going on. Lot’s of possible causes to this issue!
My hunch is that you have zero-variance balances because you are including “subject” in your formula. What happens if you exclude “subject”? But if that does not work, please check the other possible causes in that string before posting back — other issues, e.g., empty features, could be the cause.
Please also make sure you are running the most recent release of QIIME2 (2018.2).
Thanks! Good luck!
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.