Training and Sharing a MaarjAM-Based Naive Bayes Classifier for AMF Taxonomic Classification in QIIME 2

Community Contributions Data resources
, ,
1

Hello QIIME 2 Community,

We would like to share a MaarjAM-based Naive Bayes taxonomic classifier developed for the identification and taxonomic classification of Arbuscular Mycorrhizal Fungi (AMF) using QIIME 2.

This resource is part of our AMF bioinformatics pipeline and is intended to support researchers working on AMF metabarcoding, fungal biodiversity, and reference-based taxonomic classification.

Along with the trained classifier, we are providing the reference sequence and corresponding taxonomy artifacts used during classifier training. This allows researchers to directly use the classifier or reproduce the training procedure using the supplied reference resources.

1. Reference database and training resources

The classifier was trained using two MaarjAM reference

maarjam_classifier.qza (4.4 MB)

Reference taxonomy:

ref-taxonomy-maarjam.qza (179.2 KB)

Contains the corresponding taxonomic information associated with the reference sequences

Reference sequences:

maarjam-ref-seq.qza (969.6 KB)

Contains the reference sequences used for classifier training..

2. Training the Naive Bayes classifier

The classifier was trained using the QIIME 2 feature-classifier plugin and its fit_classifier_naive_bayes method.

The reference sequence and taxonomy artifacts were supplied as inputs:

from qiime2.plugins.feature_classifier.actions import fit_classifier_naive_bayes

maarjam_classifier = fit_classifier_naive_bayes(
    reference_reads=maarjam_ref_seq,
    reference_taxonomy=ref_taxonomy_maarjam
).classifier

This generates a trained taxonomic classifier artifact.

3. Trained classifier

The resulting classifier is saved as a QIIME 2 artifact:

maarjam_classifier.qza

This trained classifier can be used for taxonomic assignment of candidate AMF representative sequences without requiring users to repeat the classifier-training procedure.

4. Applying the classifier

The trained classifier can be applied to candidate AMF sequences using the QIIME 2 classify_sklearn method.

For example:

taxonomy_maarjam_sklearn, = (
    feature_classifier_actions.classify_sklearn(
        reads=amf_candidate_rep_seqs,
        classifier=maarjam_classifier,
        n_jobs=24
    )
)

The resulting taxonomic assignments can be saved as QIIME 2 artifacts and examined using QIIME 2 visualization tools.

5. Available resources

The following QIIME 2 artifacts are included in the resource collection:

File Description
maarjam-ref-seq.qza MaarjAM reference sequences
ref-taxonomy-maarjam.qza Corresponding reference taxonomy
maarjam_classifier.qza Trained Naive Bayes classifier

Providing the trained classifier alongside its reference inputs supports both direct application and reproducibility of the classifier-training procedure.

6. Software compatibility

The associated workflow was developed using:

  • QIIME 2 framework: 2025.4.0

  • Python: 3.10.14

  • scikit-learn: 1.4.2 (version reported during classifier training)

Compatibility note: During classifier training, QIIME 2 generated a warning indicating that the classifier was trained using scikit-learn 1.4.2 and may produce unreliable results when used with other scikit-learn versions.

Users should therefore maintain a compatible software environment when applying this classifier.

7. Resource access

The trained classifier, reference sequence artifact, and reference taxonomy artifact are available through our Zenodo preview record:

INNOVATIVE PIPELINE FOR ACCURATE AMF QUANTIFICATION

Access the Zenodo preview record

8. Community discussion

We welcome feedback and suggestions from the QIIME 2 community regarding:

  • Application of custom MaarjAM-based classifiers for AMF identification.

  • Training taxonomic classifiers using reference sequence and taxonomy artifacts.

  • Compatibility and reproducibility of trained QIIME 2 classifiers.

  • Evaluation of AMF taxonomic assignments using mock-community datasets.

  • Potential improvements to AMF-specific taxonomic classification workflows.

We hope that sharing the trained classifier and its reference inputs will be useful to researchers working on AMF metabarcoding and fungal biodiversity analysis.

Thank you to the QIIME 2 community for providing a platform for sharing reproducible bioinformatics resources.