Ion Torrent Proton metagenome sequences : no results

[smatamoros]

Hello,

I hope someone here can help with this rather weird problem we are having.

For a recent project, we seqenced the metagenome from several patients using the IonTorrent Proton plateform. The results obtained while analyzing these sequences with Metaphlan2 are quite puzzling.

The only identification that Metaphlan provides is some unnamed viruses... kViruses|pViruses_noname|cViruses_noname|oViruses_noname|fViruses_noname|gViruses_noname|sChilli_leaf_curl_Multan_alphasatellite|tPRJNA39933

I know my samples have a high human DNA contamination level, so I removed most of them using Deconseq. However that didn't improve the results, I still have the same weird viruses as only results.

Is there a known issue of Metaphlan with Proton sequences ? Or is there just too many human sequences in my sample ?

Thanks !

[maxchatka]

Hey Smatamoros,

we are on the same boat!! I run my NextSeq reads from soy sauce samples and got 98% of sequences to viral DNA as well. There is definitely issue with default parameters of metaphlan2 pipeline. Unfortunately, we have not identified the issue yet, so i only can assure you that the output of metaphlan2 is not correct. Metaphlan2 can found viruses in any sample, so be careful with interpretation.

I hope maybe someone can help us to explain on this issue??

[smatamoros]

Hi,

Yes, we still have this problem, however there are a few tricks to work around it.

As far as we understand, the high virus content does not represent the actual number of sequences, but the weighted output provided by Metaphlan, which takes into account the genome size : viruses have very small genomes, so the balance is off. My guess is that from a soy sauce sample you are picking a lot of soy and fish DNA, which the program mistakenly identifies as viruses, due to the similarities between portion of eukaryotic genomes and ancient viruses (CRISPR sequences...).

You can run Metaphlan without viruses or eukaryotic databases using the options :

--ignore_viruses  --ignore_eukaryotes

The --stat and -t options are also useful to get a more detailed overview of your results.

For example

--stat avg_g

and

--stat avg_l

are useful to see the bacterial clades that are outside of the quantile limit in the standard settings.

[bastianwur]

For another quick check use Kraken, which does the assignment based on kmer comparison to a database.
It comes with a bacterial database, I think, and is relatively fast.

[mlirski]

a hint

Check out read length histogram. It's possibble that it's a result of adapter/primer dimer contamination. All top blast hits for PI adapter sequence are viruses...