BUG: ANCOM-BC Fails if sample IDs look too much like exponent

[cherman2]

Bug Description We had a user on the forum that reported getting the error message "duplicate 'row.names' are not allowed " from ANCOM-BC. For this user it was because their sample-ids looked like exponents. My theory is that the sample-ids are losing some resolution when they are converted to exponents and so they are becoming "duplicates". I was able to fix this by replacing sample-ids with E to F and then it worked fine.

Forum x-ref

Steps to reproduce the behavior

1. Run ancom-bc with forum users data

Steps to reproduce the solution In python:

1. from qiime2 import Artifact
2. import pandas as pd
3. table = Artifact.load("/Users/chloeherman/Downloads/forum-table.qza")
4. df = table.view(pd.DataFrame)
5. df.to_csv("/Users/chloeherman/Downloads/forum-table-3.txt", sep = "\t", header = True)
6. edit ids in txt to not have "E"
7. Go to R
8. run read.delim(file, check.names = FALSE, row.names = 1) < this will fail if you haven't done step 6.

[colinbrislawn]

I'm interested in helping with this one.

[gregcaporaso]

Thanks @colinbrislawn, that would be great. Do you have an idea of timeline? We do have a release coming up around mid-Feb, and it would be nice to get a fix in for this by then (so would probably need a PR by around the end of next week for that to happen).

Let us know if you need any input.

[colinbrislawn]

If you can provide some support, I bet I can get this submitted by next week.

[gregcaporaso]

Sounds good @colinbrislawn, what kind of support would you need? Someone on the team in my lab should be able to help out as needed.

[colinbrislawn]

what kind of support would you need?

idk, let's find out!

Chloe, can you send me the forum-table.qza or forum-table-3.txt, Or maybe just post head forum-table-3.txt here?

I hope this can be fixed with a careful use of read.delim read_csv read_tsv to better support sample names.

[colinbrislawn]

I've confirmed what Chloe suspected, which is that specific sample names are being converted to numbers.

This only happens when all sample name look like SI notation, e.g. 0001E002 If a single name breaks the format, like L0001E002, it's fine, which is why we haven't see it sooner.

In my testing, simply adding colClasses = c("character") to read.delim() fixes it. I've opened a PR: #131

More testing is needed!

[cherman2]

Thanks for looking into this. Looks like you made some progress so you might not need it but here is forum-table-2.txt forum-table-2.txt This is the file that I got to pass by changing the Es in the sample id column to Fs

[colinbrislawn]

Proposed change log:

q2-composition

@colinbrislawn and @lizgehret fixed a :bug: in ANCOM-BC where sample IDs that look like scientific notation (1e23) where being interpreted as numbers.

[lizgehret]

changelog text looks great, thanks @colinbrislawn! will get this added in the next couple of weeks 🙂