# Asking for some details regarding the feature classifier - training for marker gene database

**URL:** <https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528>\
**Category:** User Support\
**Tags:** feature-classifier\
**Created:** [February 25, 2019, 9:26pm UTC](https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528 "2019-02-25T21:26:44Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![mstagliamonte](https://forum.qiime2.org/user_avatar/forum.qiime2.org/mstagliamonte/32/4373_2.png) [@mstagliamonte](https://forum.qiime2.org/u/mstagliamonte)\
**Post date:** [February 25, 2019, 9:26pm UTC](https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528/1 "2019-02-25T21:26:44Z")

</div>

Dear Qiimers,

I am trying to set up a database and training the classifier for a marker gene other than 16S. I have the primers set and a fasta file; the sequences are currently very different lengths (500 to 5,000 nts), and I am not sure there will be duplicates after the step 'extract reads' step. I have a couple of questions regarding it:

1. I did a test run using the GreenGenes database, and I have noticed that after the 'extract-reads', the number of reference sequences is about half or the original set (99\_otus.fasta). How does it work? Does the algorithm merges multiple sequences if they become identical (after extracting the intra-primers segment)? And then merges the taxonomic assignment to the lowest common classification?

2. Should I trim my reference sequences to length, then cluster them at a given similarity percentage before generating the database? At this point I am planning to run Deblur (or DADA2) on my sequence data, then use the taxonomic assignment just as a rough guide.

Hope this makes sense, please let me know if you need additional details.

Thank you for your kind attention,  
Max

---

<div class="post-metadata">

**Author:** ![Nicholas\_Bokulich](https://forum.qiime2.org/user_avatar/forum.qiime2.org/nicholas_bokulich/32/19937_2.png) [@Nicholas\_Bokulich](https://forum.qiime2.org/u/Nicholas_Bokulich)\
**Post date:** [February 25, 2019, 10:00pm UTC](https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528/2 "2019-02-25T22:00:52Z")

</div>

> [@mstagliamonte](#):
>
> I did a test run using the GreenGenes database, and I have noticed that after the ‘extract-reads’, the number of reference sequences is about half or the original set (99\_otus.fasta).

Sequences that do not match within X mismatches are dropped (see the help docs — there is a parameter to adjust the % mismatch tolerance). There is no dereplication going on.

> [@mstagliamonte](#):
>
> Should I trim my reference sequences to length,

See the notes in the feature classifier training tutorial. It helps, but it is not essential.

> [@mstagliamonte](#):
>
> then cluster them at a given similarity percentage before generating the database?

That is not necessary, though it would make some steps more efficient. As yet, we have no best-practices guide for how to do this, nor can that clustering be achieved using QIIME 2 (you can cluster the sequences but there is no way to create a consensus taxonomy from those clusters).

Good luck!

---

<div class="post-metadata">

**Author:** ![mstagliamonte](https://forum.qiime2.org/user_avatar/forum.qiime2.org/mstagliamonte/32/4373_2.png) [@mstagliamonte](https://forum.qiime2.org/u/mstagliamonte)\
**Post date:** [February 26, 2019, 2:04pm UTC](https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528/3 "2019-02-26T14:04:27Z")

</div>

Thank you for the additional details and advice.

> Good luck!

Thanks 😉

Have a nice day,  
Max

---

<div class="post-metadata">

**Author:** ![system](https://forum-qiime2-org.s3.dualstack.us-west-2.amazonaws.com/original/3X/2/1/21af5fe23cb6f4579467c66a9ed94e55274ca7bd.svg) [@system](https://forum.qiime2.org/u/system)\
**Post date:** [March 29, 2019, 8:04pm UTC](https://forum.qiime2.org/t/asking-for-some-details-regarding-the-feature-classifier-training-for-marker-gene-database/8528/4 "2019-03-29T20:04:32Z")

</div>

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