Eukaryote CO1 Reference Set For The Identification of DNA Metabarcodes

This repository contains training sets that can be used with the Ribosomal Database Project classifier (Wang et al., 2007) to taxonomically assign Eukaryote CO1 mtDNA sequences. The latest release can be downloaded from https://github.com/terrimporter/CO1Classifier/releases . The trained files ready to be used with the RDP Classifier are available as well as the original files used for training (a taxonomy file and a FASTA file) are available as 'version-ref'.

**As of version 5.0.0, reference sequences formatted for the RDP classifier, BLAST, and SINTAX are also available.**

Quick start

############ Install the RDP classifier if you need it
# The easiest way to install the RDP classifier v2.13 is using conda
conda install -c bioconda rdp_classifier
# Alternatively, you can install from SourceForge and run with java if you don't use conda
wget https://sourceforge.net/projects/rdp-classifier/files/rdp-classifier/rdp_classifier_2.13.zip
# decompress it
unzip rdp_classifier_2.13
# record path to classifier.jar ex. /path/to/rdp_classifier_2.13/dist/classifier.jar

############ Get the latest COI training set
wget https://github.com/terrimporter/CO1Classifier/releases/download/RDP-COI-v5.1.0/RDP_COIv5.1.0.zip

# decompress it
unzip RDP_COIv5.1.0.zip

# record the path to the rRNAClassifier.properties file ex. /path/to/mydata_trained/rRNAClassifier.properties

############ Run the RDP Classifier 
# If it was installed using conda, run it like this:
rdp_classifier -Xmx8g classify -t /path/to/mydata_trained/rRNAClassifier.properties -o rdp.output query.fasta
# Otherwise run it using java like this:
java -Xmx8g -jar /path/to/rdp_classifier_2.13/classifier.jar -t /path/to/mydata_trained/rRNAClassifier.properties -o rdp.output query.fasta

How to cite

If you use these training sets in a publication, please cite:

CO1 Classifier publication

Porter, T.M., & Hajibabaei, M. (2018) Automated high throughput animal CO1 metabarcode classification. Scientific Reports, 8, 4226.

CO1 Classifier code repository

Teresita M. Porter. (2017, December 4). Eukaryote CO1 Reference Set For The RDP Classifier (Version v4.0.1). Zenodo. http://doi.org/10.5281/zenodo.4741447

The RDP classifier

Wang et al. (2007) Naïve Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and Environmental Microbiology, 73: 5261.

Releases

v5.1.0 ** NEW **

This version was updated to:

1. Fix a problem with commas present within result fields causing misalignment of columns.
2. Added a step to consolidate BIN IDs in GenBank records with newer Linnean binomial species names (if present) from BOLD data package BOLD_Public.24-Mar-2023 .

This version includes eukaryote COI sequences mined from GenBank [February 2023]. This version includes records deposited between 1982 - 2022 (inclusive). GenBank sequences were filtered to only include sequences 500bp+, containing no nucleotide ambiguities, and preferrably with a Linnean binomial species name and/or a BOLD BIN (new). Sequences were screened to remove human and bacterial contaminants. Bacterial outgroup sequences were added. Human sequences are included.

This version is based on 2,216,547 COI sequences from 236,247 taxa including 185,389 species/BINs. This is an increase of 995,019 more sequences, 81,896 more taxa, and 70,693 more species compared to v4.

Accuracy was assessed using 5-fold cross validation (new). The metazoan sequences in the classifier were divided into 5 groups. For each fold of the data, the remaining four folds were combined (i.e., 20% test, 80% train) with the outgroup sequences to create a training set. The average number of correctly classified metazoan sequences from each fold were used to calculate the cutoffs below.

Assuming that your query sequences are present in the reference set, using these minimum bootstrap support cutoffs should result in at least 99% correct assignments:

Rank | 500bp+ | 400 bp | 300 bp | 200 bp | 100 bp --- |:---:|:---:|:---:|:---:|:---:
Superkingdom | 0 | 0 | 0 | 0 | 0
Kingdom | 0 | 0 | 0 | 0 | 0
Phylum | 0 | 0 | 0 | 0 | 0
Class | 0 | 0 | 0 | 0 | 0
Order | 0 | 0 | 0 | 0 | 0.1
Family | 0 | 0 | 0 | 0.1 | 0.2
Genus | 0.3 | 0.3 | 0.3 | 0.3 | 0.3
Species | NA | NA | NA | NA | NA

NA = No cutoff available will result in 99% correct assignments

If you really want to work with species level data, then assuming that your query sequences are present in the reference set, using these minimum bootstrap support cutoffs should result in at least 95% correct assignments:

Rank | 500bp+ | 400 bp | 300 bp | 200 bp | 100 bp
--- |:---:|:---:|:---:|:---:|:---:
Superkingdom | 0 | 0 | 0 | 0 | 0
Kingdom | 0 | 0 | 0 | 0 | 0
Phylum | 0 | 0 | 0 | 0 | 0
Class | 0 | 0 | 0 | 0 | 0
Order | 0 | 0 | 0 | 0 | 0
Family | 0 | 0 | 0 | 0 | 0
Genus | 0 | 0 | 0 | 0 | 0
Species | 0.9 | 0.8 | 0.8 | 0.8 | 0.8

If you are okay with more lenient cutoffs, as this method has been shown to have a high false negative rate in practice, then using these minimum bootstrap support cutoffs should result in at least 90% correct assignments:

Rank | 500bp+ | 400 bp | 300 bp | 200 bp | 100 bp
--- |:---:|:---:|:---:|:---:|:---:
Superkingdom | 0 | 0 | 0 | 0 | 0
Kingdom | 0 | 0 | 0 | 0 | 0
Phylum | 0 | 0 | 0 | 0 | 0
Class | 0 | 0 | 0 | 0 | 0
Order | 0 | 0 | 0 | 0 | 0
Family | 0 | 0 | 0 | 0 | 0
Genus | 0 | 0 | 0 | 0 | 0 Species | 0 | 0 | 0 | 0.1 | 0.2

v5

This version was updated to include eukaryote COI sequences mined from GenBank [February 2023]. This version includes records deposited between 1982 - 2022 (inclusive). GenBank sequences were filtered to only include sequences 500bp+, containing no nucleotide ambiguities, and preferrably with a Linnean binomial species name and/or a BOLD BIN (new). Sequences were screened to remove human and bacterial contaminants. Bacterial outgroup sequences were added. Human sequences are included.

**This version is based on 2,211,191 COI sequences from 239,382 taxa including 188,534 species/BINs. This is an increase of 989,663 more sequences, 85,031 more taxa, and 73,847 more species compared to v4.**

Accuracy was assessed using 5-fold cross validation (new). The metazoan sequences in the classifier were divided into 5 groups. For each fold of the data, the remaining four folds were combined (i.e., 20% test, 80% train) with the outgroup sequences to create a training set. The average number of correctly classified metazoan sequences from each fold were used to calculate the cutoffs below.

[Cutoff values will NOT be added, please use v5.1.0 instead]

v4

This version was updated to include COI sequences mined from GenBank [April 2019] and the BOLD data releases [iBOL_phase2.0_COI.tsv to iBOL_phase_6.50_COI.tsv] from http://v3.boldsystems.org/index.php/datarelease . GenBank sequences were filtered to only include those 500bp+, containing no nucleotide ambiguities, and with a Linnean binomial species name. BOLD sequences were filtered to only retain sequences 500bp+, with less than 3 ambiguous bases, and containing a Linnean binomial species name. Sequences were screened for human and bacterial contaminants. Bacterial outgroup sequences were added.

This page will be updated with updated cutoffs when available. For now use the recommended cutoffs for v3 (below).

v3.2

This version was updated to include some invasives species of interest even though their sequences are less than 500bp in length.

The latest release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.2 The CO1v3_2_trained.tar.gz file should be decompressed and used directly with the RDP Classifier to make taxonomic assignments to the species rank.

The reference files for the latest release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.2-ref The CO1v3_2_training.tar.gz file should be cecompressed. The folder contains the original taxonomy and fasta files that are included here for reference only. They are the same as the v3 files except that a few bacterial outgroup sequences that were misannotated as bacteriophage viruses were removed.

The v3 MINIMUM bootstrap cutoff values should be used with this version of the CO1 classifier.

v3.1

The v3.1 release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.1 The CO1v3_1_trained.tar.gz file should be decompressed and used directly with the RDP Classifier to make taxonomic assignments to the species rank.

The reference files for the v3.1 release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.1-ref The CO1v3_1_training.tar.gz file should be cecompressed. The folder contains the original taxonomy and fasta files that are included here for reference only. They are the same as the v3 files except that a few bacterial outgroup sequences that were misannotated as bacteriophage viruses were removed.

The v3 MINIMUM bootstrap cutoff values should be used with this version of the CO1 classifier.

v3

The v3 release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.0 The CO1v3_trained.tar.gz file should be decompressed and used directly with the RDP Classifier to make taxonomic assignments to the species rank.

The reference files for the v3 release can be downloaded from here: https://github.com/terrimporter/CO1Classifier/releases/tag/v3.0-ref The CO1v3_training.tar.gz file should be decompressed. The folder contains the original taxonomy and fasta files that are included here for reference only. They were originally mined from GenBank in April 2018. These seque