# How to train the classifier with multiple reverse primers?
**URL:** https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581
**Category:** User Support
**Tags:** feature-classifier
**Created:** [March 28, 2018, 2:43pm UTC](https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581 "2018-03-28T14:43:48Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@wym199633](https://forum.qiime2.org/u/wym199633)
#### Post date: [March 28, 2018, 2:43pm UTC](https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581/1 "2018-03-28T14:43:48Z")
Hi everyone,
A got the sequencing data from elsewhere, and the author said that
"The 16S-specific primers contain degenerate sites or, in the case of 926R, represent a combination of three distinct oligonucleotides in order to capture broad eubacterial diversity."
This is the forward and the reverse primer:
|518F|CCAGCAGCYGCGGTAAN |v4-v5 forward primer|
|926R1|CCGTCAATTCNTTTRAGT |v4-v5 reverse primer|
|926R3|CCGTCAATTTCTTTGAGT |v4-v5 reverse primer|
|926R4|CCGTCTATTCCTTTGANT |v4-v5 reverse primer|
As it shows, there are three reverse primers they used. I want to use the GreenGenes 16S reference sequences. Is it ok to only choose one reverse primer for extract reference reads? According to [Training feature classifiers with q2-feature-classifier — QIIME 2 2018.2.0 documentation](https://docs.qiime2.org/2018.2/tutorials/feature-classifier/#extract-reference-reads), it seems I can only use one reverse primer....But will it bring any bias?
---
### Author:  [@colinbrislawn](https://forum.qiime2.org/u/colinbrislawn)
#### Post date: [March 28, 2018, 6:47pm UTC](https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581/2 "2018-03-28T18:47:38Z")
Good morning,
Using only a single primer primer could introduce bias, but fortunately there are other settings for [extract reads](https://docs.qiime2.org/2018.2/plugins/available/feature-classifier/extract-reads/) that could help mitigate this issue. For example
```bash
--p-identity FLOAT minimum combined primer match
identity threshold. [default: 0.8]
```
Let's see how similar your three reverse primers are!
```auto
|926R1|CCGTCAATTCNTTTRAGT |v4-v5 reverse primer|
|926R3|CCGTCAATTTCTTTGAGT |v4-v5 reverse primer|
|926R4|CCGTCTATTCCTTTGANT |v4-v5 reverse primer|
926R1 vs 926R3 == 3 differences
926R3 vs 926R4 == 3 differences
926R1 vs 926R4 == 4 differences
```
So now when we look at the `p-identity` default in the script (.8), we could compare this to the three differences in the script.
```auto
3 differences / 18 bp length == 83% similar
80% similar over 18 bp == less than or equal to 3.6 bp different
```
So you could reasonably use just the `926R3` primer, and the default `p-identity` will let you capture matches to the other two primers. If you want to make sure you get more reads, you could use a lower identity selection. Say `--p-identity 0.7`
Let me know if this helps!
Colin
---
### Author:  [@wym199633](https://forum.qiime2.org/u/wym199633)
#### Post date: [March 28, 2018, 10:50pm UTC](https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581/3 "2018-03-28T22:50:21Z")
Hi Colin,
Thank you for the explanation! This perfectly solved my question! 😁
Yours,
Oliver
---
### Author:  [@system](https://forum.qiime2.org/u/system)
#### Post date: [April 29, 2018, 4:50am UTC](https://forum.qiime2.org/t/how-to-train-the-classifier-with-multiple-reverse-primers/3581/4 "2018-04-29T04:50:22Z")
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.