Problem with training 18S classifier and assigning taxonomy

User Support
1

Dear QIIME2 developers,
I am trying to use qiime feature-classifier to train a classifier for the use of 18S sequence.
here is the primer I used to extract the v4 target sequence:

565F CCAGCASCYGCGGTAATTCC 948R ACTTTCGTTCTTGATYRA

The confusing thing is that the length of sequences extracted by “qiime feature-classifier extract-reads” is far more than 948-565=383 (average 800 nt actually)

The even more confusing thing is that the taxonomy i got from “qiime feature-classifier classify-sklearn” is exactly all the same for all 330 features. Just for brief view:

|Feature ID|Taxonomy|
|0b298f6d9f609ae27dff8397c6b5dfea|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|e49af68b1ed288f280740f1efd0f628a|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|4e798893271f5ecce40feed01c675957|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|42613dfbaf97fdcd05e4c7da6852390d|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|581e00979f3019eb8a88ea8f05be2110|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|9c0bddb2fa5aa59e7059faefdd81b161|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|c07c010e7f93e777d631ebf52b1c3809|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|d11b4aa67389d800c12dbfbb1abe8f62|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|f1d3bed5959ea9b11ec2a8ea15e4a0ac|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|1a254c4fdce63af73175a6caba4aae73|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|e3b005dbf7d747de71d29b8ebbcb5fdb|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|2de3cb6cd76a45c01ce5daf55122c28d|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|09e9fb7ab6765be1d52adb5b4bf37a94|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|160f27c869746e32d4924e18d2049c35|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|9fc64f6b75f145c4657e3b939231c7f9|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|1715e06abf6db633d7a01ab0122c31b4|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|
|90289e6e368fd01fbeee6d7562b32228|D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis|

But if I use the full-length classifier, the taxonomy go back to normal, with different taxonomic names for every features.

here are the commands i used for training classifier:

_qiime tools import --type ‘FeatureData[Sequence]’ --input-path SILVA_132_QIIME_release/rep_set/rep_set_18S_only/99/silva_132_99_18S.fna --output-path rep-set.qza&&_
_qiime tools import --type ‘FeatureData[Taxonomy]’ --input-format HeaderlessTSVTaxonomyFormat --input-path SILVA_132_QIIME_release/taxonomy/18S_only/99/consensus_taxonomy_7_levels.txt --output-path ref-taxonomy.qza&&_
_qiime feature-classifier extract-reads --i-sequences rep-set.qza --p-f-primer CCAGCASCYGCGGTAATTCC --p-r-primer ACTTTCGTTCTTGATYRA --o-reads ref-seqs.qza&&_
qiime feature-classifier fit-classifier-naive-bayes --i-reference-reads ref-seqs.qza --i-reference-taxonomy ref-taxonomy.qza --o-classifier classifier.qza

and here is the command i used for assign taxonomy:

qiime feature-classifier classify-sklearn --p-n-jobs 16 --i-classifier classifier.qza --i-reads rep-seqs.qza --o-classification taxonomy.qza

I am using the qiime2-2018.8 version of qiime2. I wonder if this is a common situation in qiime2, or if I was wrong about something. I would appreciate that if someone would like to help me figure this out.

Thanks

3

Hi @bayegy,
Thank you for reporting this! I know just what the problem is.

This is not a common situation and you did not do anything wrong. This is a known bug that occurs when you combine certain databases with certain primer sets, resulting in some extremely short sequences. See this topic for a previous description and workaround.

The easiest thing to do would be to use the full-length 18S rRNA gene sequences, but another option is to remove the short sequences from the trimmed database (hint: D_0__Eukaryota;D_1__Opisthokonta;D_2__Holozoa;D_3__Metazoa (Animalia);D_8__Arachnida;D_9__Opiliones;D_10__Karamea lobata australis will be one of them! But there might be one or two others).

We have an issue to track this. This is on our radar to get fixed in the near future, so stay tuned for more details!

5

Thanks for your quick and detailed reply, @Nicholas_Bokulich, looking forward to this bug being fixed.

6

It should be fixed in the next release of QIIME 2 (end of this week).

1 Like
7

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.