We recently analysed some metagenomic 16S data using QIIME (v1.9.1) and the SILVA database (release 128) for taxonomy assignment and then fed these results into LefSe (Segata et al., 2011 - http://doi.org/10.1186/gb-2011-12-6-r60) (we recently installed it locally with docker) to determine which clades were significantly more present in one group of samples than the other.
We understand most of the results, but there are some we find difficult to interpret when we compare different levels of analysis (ex. L2 with L5, with L7) (please, click on the question title to view the figures). For example, we see a phylum (Proteobacteria in the files attached) which is significantly enriched in treatment S (sugar) that is not then reported as significantly increased in the L5 analysis and then appears again (in isolation) in the L7 level . We understand there is no specific family within the Proteobacteria that is enriched, but there are some species that make the difference, hence the Proteobacteria bar appearing again in L7. We hope this is the right interpretation...
But when the opposite happens, we are at a loss for explanations: see for example family Lachnospiraceae appearing as a significantly enriched clade in treatment "Control" in L7, but being absent in the L5 results of the same analysis: how can this be explained? If family Lachnospiraceae is found as significant at the L7 level, why doesn't it appear at the L5 (family) level using the same LDA cutoff?
Can anybody help us explain this?
We would really appreciate.
Thank you very much Nsa Dada ,
Indeed, after some tests, we believe that the problem may be in the input file, more precisely, when we generate the L2-L7 tables from the biom file in QIIME with summarize_taxa.py script...
Thank you Sergei Drovetski
Indeed, we think this would be a good idea; however, we don’t know how we can use the biom file with all the information directly with LEfSe… furthermore, we want to use “Treatment” as a class, but in our biom file we don’t have the information about the treatment; this information is provided in a mapping file. We usually merge these data when we create the L2-L7 tables in QIIME with summarize_taxa.py, and then we use each of these tables as input for LEfSe. If we use use the biom file directly in with LEfSe, how can we use the information about the treatment? Do you know how this is possible? Thank you very much!
Hi Sergei and Nsa, this is Paolo, Ana's colleague. Thank you both for your valuable suggestions.
Unfortunately, after running some test files with LefSe, we come to the conclusion that the program has a big flaw: it isn't capable of treating as a group all the lower taxa underneath the name of a higher taxon in its analysis. For instance, a significant result for k__Bacteria; p__Firmicutes; c__Clostridia; o__Clostridiales; f__Lachnospiraceae in treatment1 over treatment2 only refers to those OTUs whose classification ends with "Lachnospiraceae", all OTUs classified at a deeper level (ex. k__Bacteria; p__Firmicutes; c__Clostridia; o__Clostridiales; f__Lachnospiraceae; g__Coprococcus; s__ or k__Bacteria; p__Firmicutes; c__Clostridia; o__Clostridiales; f__Lachnospiraceae; g__Eubacterium; s__Eubacterium_ventriosum) are not taken into account in the Lachnospiraceae result. (Obviously, this also happens with other families and other taxonomic levels)
Even when we run LefSe at discrete levels (in this example family level L5) the Lachnospiraceae result refers only to those OTUs whose classification is truncated at the family level. One would think that summarize_taxa at L5 level should delete from each OTU name all genus and species info and then LefSe should work with the resulting truncated names, treating all OTUs now ending with "Lachnospiraceae" as a group, but our test shows it doesn't work this way... I guess summarize_taxa keeps only those OTUs with classification ending at the family level...
I guess we could manually truncate all the OTUs names ourselves at the level we want to analyse (ex. family - was it this Nsa was suggesting?), but with 65,000 lines in our real data file, this becomes unmanageable... Any more ideas to help us with this issue?
Thank you again.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts