But when I run the basic analysis as a QIIME2 plugin, it only generates the conditional_biplot.qza and conditionals.qza files, NOT the model_stats.qza file. Code below:
I've been searching through the forum and the package documentation, and cannot find anyone with a similar issue. Is this a problem?
I am using QIIME2/2023.5, and mmvec/1.0.5, and I have attached my mmvec output files below, as well as the qiime artifacts I used to run the analysis through the plugin.
Hi @ceraker, Welcome to the QIIME 2 Forum! I ping'ed the developer of the mmvec plugin to help out with this one, so we should have some input for you soon.
No, I'm not sure why it is the case -- I wonder if you didn't train your model long enough (i.e. it completed within 1 second). It logs every second, so if you train too fast, that could lead to no logged outputs.
And my model definitely took longer than 1 second, but shorter than five minutes, so it is possible that I did not train it for long enough. I will change some parameters and get back to you.
RE ordination - you can't visualization 1D PCoAs. You need at least 2 dimensions for that.
Regarding why no model_stats is outputted, I'm really not sure -- this is the first time this error has occurred. The only thing that could explain this is that something with the time readouts is broken (which would be bizarre).
Have you tried the non-qiime2 version? It would be another way to sanity check it.
Feel free to provide the raw readouts. I won't have time to look at this in the near future, but will try to look when time frees up.
Okay, on ordination: I thought I HAD more than two dimensions! If I don't, then I'm not sure what's going on with my data.
As for install issues, we (my university's computing office and myself) had issues installing the qiime2 plugin version, and were finally able to make it work in a container. So that could be part of it? I have never worked with containers before, so I'm not very familiar with troubleshooting.
And I have run the non-qiime2 version, which outputted the ranks, ordination, and embedding files I attached in my initial post.
Thank you so much for looking into this, I really appreciate it!
Sorry, back again. I am attempting to essentially re-do this analysis from the beginning, and something seems to have gone wrong with my input files. My code to convert things to the .biom format and incorporate metadata runs fine, but then when I try to run a basic mmvec paired-omics analysis I get the error: ValueError: 'a' cannot be empty unless no samples are taken. I'm really not sure why this is happening!
For reference, here is the code I am using to create the metabolome .biom file:
module load mmvec/1.0.5
Apologies if I'm doing something very silly with the file conversions: I've tried a couple different methods (particularly with the microbiome data, which is from qiime), but continue getting the same error. Input files are attached. table-filtered_blue.qzv (3.4 MB) meta_samp_met.txt (8.7 KB) metabolites.txt (841.0 KB) met_obs_meta.txt (53.5 KB)
Back again now that everything is installed properly (hooray!) and I can finally get back to actually analyzing my data.
I have metabolomic data from an outside source, and microbiome data from qiime2 v 2023.5, and am now trying to analyze it using the qiime2 2020.6 mmvec plugin. I am running into issues with I believe my microbiome data.
I have tried multiple ways to convert it and upload it, but when I run the analysis I get the error:
Plugin error from mmvec:
'a' cannot be empty unless no samples are taken
I get the same error when I run mmvec on its own (not as a plugin). I've searched various forums, but continue to be baffled. I definitely have samples, and looking at the example data from the Knight lab, mine seems to be formatted the same way.
I've attached a link to the files I've been using, and the code to convert things to biom below:
Hi, sorry about following up late. I recommend checking against or posting on the MMvec github issue if you haven't already solved this.
There are two issues posted below that often points to issues with the input data (where all of the samples get filtered out due to mismatch in IDs or sampling depth).