Taxonomic Analysis

Last updated on 2026-10-06 | Edit this page

Estimated time: 12 minutes

Overview

Questions

  • How is taxonomy assigned to the representative sequences?
  • Why must the classifier match the amplified region?
  • What filtering steps are commonly applied to taxonomic results?

Objectives

  • Apply a pre‑trained classifier appropriate for the V4 16S region to assign taxonomy to ASVs.
  • Generate a taxonomic summary visualization and interpret broad community composition.
  • Filter out non‑target classifications (mitochondria, chloroplast) from the feature table.

Assign taxonomy

Here we will classify each identical read or Amplicon Sequence Variant (ASV) to the highest resolution based on a database. Common databases for bacteria datasets are SILVA, Ribosomal Database Project*, or Genome Taxonomy Database. See Porter and Hajibabaei, 2020 for a review of different classifiers for metabarcoding research. The classifier chosen is dependent upon:

  1. Previously published data in a field
  2. The target region of interest
  3. The number of reference sequences for your organism in the database and how recently that database was updated.

A classifier has already been trained for you for the V4 region of the bacterial 16S rRNA gene using the SILVA database. The next step will take a while to run. The output directory cannot previously exist.

n_jobs = 1 This runs the script using all available cores

As of the time of writing, the Ribosomal Data Project website is no longer available. You can find a standalone version of the RDP Classifier 2.14 released in August 2023 on Sourgeforce and Zenodo.

Callout

The classifier used here is only appropriate for the specific 16S rRNA region that this data represents. You will need to train your own classifier for your own data. For more information about training your own classifier, see Extra Information.

BASH 

qiime feature-classifier classify-sklearn \
--i-classifier silva_138.2_16s_v4_classifier.qza \
--i-reads analysis/dada2out/representative_sequences.qza \
--p-n-jobs 1 \
--output-dir analysis/taxonomy \
--verbose
Caution

This step often runs out of memory on full datasets. Some options are to change the number of cores you are using (adjust --p-n-jobs) or add --p-reads-per-batch 10000 and try again. The QIIME 2 forum has many threads regarding this issue so always check there was well.

Generate a viewable summary file of the taxonomic assignments.

BASH 

qiime metadata tabulate \
--m-input-file analysis/taxonomy/classification.qza \
--o-visualization analysis/visualisations/taxonomy.qzv \
--verbose
Challenge

Visualisation: Taxonomy

Copy analysis/visualisations/taxonomy.qzv to your local computer and view in QIIME 2 View (q2view).

Filtering

Filter out reads classified as mitochondria and chloroplast. Unassigned ASVs are retained. Generate a viewable summary file of the new table to see the effect of filtering.

According to QIIME developer Nicholas Bokulich, low abundance filtering (i.e. removing ASVs containing very few sequences) is not necessary under the ASV model.

BASH 

qiime taxa filter-table \
--i-table analysis/dada2out/table.qza \
--i-taxonomy analysis/taxonomy/classification.qza  \
--p-exclude Mitochondria,Chloroplast \
--o-filtered-table analysis/taxonomy/16s_table_filtered.qza \
--verbose

BASH 

qiime feature-table summarize \
--i-table analysis/taxonomy/16s_table_filtered.qza \
--m-metadata-file dunnart_metadata.tsv \
--output-dir analysis/visualisations/summary_table_filtered \
--verbose
Challenge

Visualisation: Feature/ASV summary-Filtered

Copy analysis/visualisations/summary_table_filtered/summary.qzv to your local computer and view in QIIME 2 View (q2view).

Key Points
  • Classifier must be trained for the same primer/region used in the dataset; mismatched classifiers give poor results.
  • The classify‑sklearn step can be memory intensive — precomputed classifications are provided for the workshop.
  • Retain unassigned ASVs (they may still be biologically meaningful), but exclude obvious host/plant contaminants.
  • Low‑abundance filtering is generally unnecessary with the ASV paradigm unless specific reasons exist.