Difference between Unite dynamic and not dynamic classifier

General Discussion
, ,
1

Hello everybody,

I'm new here and I’d like to know if anyone could help me by clarifying a doubt about the Unite fungal classifier. I read some topics before but I still have some doubts.

What is the difference between using the “normal’ Unite classifier and the dynamic classifier? Because when I performed the taxonomy training by importing sh_refs_qiime_ver7_dynamic_01.12.2017.fasta and then tested importing sh_refs_qiime_ver7_97_01.12.2017.fasta, I saw a difference in the final result. Ex: in one classification I found Microsporum canis and in another Microsporum audouinii . These were the commands that I used:

Thank you in advance for this kind community!

qiime tools import
--type 'FeatureData[Sequence]'
--input-path sh_refs_qiime_ver7_dynamic_01.12.2017.fasta
--output-path unite1.qza

qiime tools import
--type 'FeatureData[Taxonomy]'
--input-format HeaderlessTSVTaxonomyFormat
--input-path sh_refs_qiime_ver7_dynamic_01.12.2017.txt
--output-path unite-taxonomy1.qza

qiime feature-classifier fit-classifier-naive-bayes
--i-reference-reads unite1.qza
--i-reference-taxonomy unite-taxonomy1.qza
--o-classifier classifier1.qza

———

qiime tools import
--type 'FeatureData[Sequence]'
--input-path sh_refs_qiime_ver7_99_01.12.2017.fasta
--output-path unite2.qza

qiime tools import
--type 'FeatureData[Taxonomy]'
--input-format HeaderlessTSVTaxonomyFormat
--input-path sh_refs_qiime_ver7_99_01.12.2017.txt
--output-path unite-taxonomy2.qza

qiime feature-classifier fit-classifier-naive-bayes
--i-reference-reads unite2.qza
--i-reference-taxonomy unite-taxonomy2.qza
--o-classifier classifier2.qza

2

Welcome @vetalinesantana!

You may want to consult the UNITE website and article for more details, but my understanding is that the "dynamic" classifier clusters different species at different % similarity thresholds, based on manual curation by taxonomic experts, as opposed to a single % identity to define species.

So it is not unusual or unexpected that dynamic would yield different results from the 99% OTUs. The question you need to ask yourself is: which do you trust? Do you find M. canis or M. audouinii more believable?

Short of having ground truth data that you can test, you might not be able to determine which is more accurate. We have actually tested these databases recently, and the clustering threshold had less impact than other factors... in fact, 99% and dynamic yielded very similar performance... so I cannot say that one stands out as more accurate.

Good luck deciding!

2 Likes
3

Hey Nicholas,

Thank you so much for your kind help.

You brought up VERY interesting information and I'll go deeper on this!

Again, thank you for your time in answering my post! It helped me a lot.

1 Like