argnorm 1.0.0

Normalize antibiotic resistance genes (ARGs) results by using the ARO ontology (developed by CARD).

argNorm: Normalize ARG annotations to the ARO

argNorm Overview

General overview of argNorm. (a) Genomes and metagenomes can be annotated using ARG annotation tools. argNorm accepts the outputs of these ARG annotation tools directly or after the outputs are processed by hAMRonization to perform ARO normalization and drug categorization. (b) The argNorm workflow includes: mapping gene names in the ARG annotation outputs to ARO accessions from ARO annotation tables constructed using RGI and manual curation; and mapping gene ARO accessions to drugs and drugs classes. In categorizing drugs, argNorm reports the immediate child of the ‘antibiotic molecule’ node. Tobramycin is an example of a drug to which the ANT(2'')-Ia gene confers resistance to and tobramycin can be categorized as an aminoglycoside antibiotic. The ‘confers_resistance_to_antibiotic’ and is_a relationships are used to navigate the ARO.

argNorm presented at the Microbiome Virtual International Forum

argNorm manuscript

For more high-level information on argNorm, please refer to the manuscript:

Svetlana Ugarcina Perovic, Vedanth Ramji, Hui Chong, Yiqian Duan, Finlay Maguire, Luis Pedro Coelho, argNorm: normalization of antibiotic resistance gene annotations to the Antibiotic Resistance Ontology (ARO), in Bioinformatics (2025) https://doi.org/10.1093/bioinformatics/btaf173

We also ask that you cite the manuscript if you use argNorm in your work.

argNorm workflow

Schematic illustration of argNorm workflow through a Resfinder output example: mapping gene names in the ARG annotation outputs to gene names from the ARO mapped ARG databases and adding corresponding drug categorization, namely “confers resistance to immediate drug class” and “overall category of drug class”, from the ARO ontology file.

What is argNorm?

argNorm is a tool to normalize antibiotic resistance genes (ARGs) by mapping them to the antibiotic resistance ontology (ARO) created by the CARD database.

argNorm also enhances antibiotic resistance gene annotations by providing drug categorization of the drugs that antibiotic resistance genes confer resistance to.

Why argNorm?

Right now, many tools exist for annotation ARGs in genomes and metagenomes. However, each tools will have distinct output formats.

The hAMRonization package can normalize file formats, but each tool will use different names/identifiers (e.g., TetA or TETA or tet(A) or tet-A are all different ways to spell the same gene name).

For a small number of isolate genomes, a human user can manually evaluate the outputs. However, in metagenomics, especially for large-scale projects, this becomes infeasible. Thus argNorm normalizes the output vocabulary of ARG annotation tools by mapping them to the same ontology (ARO).

Besides performing normalization, argNorm also provides categorization of drugs that antibiotic resistance genes confer resistance to.

For example, the PBP2b (ARO:3003042) gene confers resistance to the drug class amoxicillin. amoxicillin is then categorized into a broader category of beta lactam antibiotic.

argNorm provides support for this, and adds the confers_resistance_to and resistance_to_drug_classes columns to ARG annotations.

The confers_resistance_to column will contain ARO numbers of all the drug classes that a gene provides resistance to (ARO:0000064 for amoxicillin in the previous example).

The resistance_to_drug_classes column will contain ARO numbers of the broader categories of the drug classes in the confers_resistance_to column (ARO:3000007 for beta lactam antibiotic in the previous example).

Supported ARG annotation tools and databases

ARG database	Tool for ARG annotation
ARG-ANNOT v5.0	ABRicate v1.0.1 & hAMRonization v1.1 and v1.0.4
DeepARG v2	DeepARG v1.0.2 & hAMRonization v1.1 and v1.0.4
Groot v1.1.2	GROOT v1.1.2 & hAMRonization v1.1 and v1.0.4
MEGARes v3.0	ABRicate v1.0.1 & hAMRonization v1.1 and v1.0.4
NCBI Reference Gene Database v3.12 and v4.0	ABRicate v1.0.1, AMRFinderPlus v3.10.30 and v4.0.19, & hAMRonization v1.1 and v1.0.4
ResFinder v4.0	ABRicate v1.0.1, ResFinder v4.0, & hAMRonization v1.1 and v1.0.4
ResFinderFG v2.0	ABRicate v1.0.1 & hAMRonization v1.1 and v1.0.4
SARG (reads mode) v3.2.1	ARGs-OAP v2.3 & hAMRonization v1.1 and v1.0.4

• Note: ARG database and ARG annotation tool versions can change. argNorm currently only supports versions listed above.
• Note: the argNorm tool will be periodically updated to support the latest versions of databases and annotation tools if they undergo significant changes.

hAMRonization: argNorm does not support ALL hAMRonized results. argNorm only supports hAMRonization outputs that were obtained by passing the outputs of the supported ARG annotation tools listed above into hAMRonization

Installation

argNorm can be installed using pip:

pip install argnorm

argNorm can also be installed through conda:

conda install bioconda::argnorm

Nextflow and funcscan

argNorm is also available as an nf-core module. If using an nf-core pipeline, argNorm can be installed using:

nf-core modules install argnorm

argNorm is readily available in the funcscan pipeline which can be accessed (here)[https://github.com/nf-core/funcscan]

Tutorial video

CLI Usage

For detailed examples of running the argNorm CLI see here

Here is a basic outline of calling argNorm.

argnorm [tool] [--db] -i [path to original_annotation.tsv] -o [path to annotation_result_with_aro.tsv] [----hamronization_skip_unsupported_tool]

Resource requirements: argNorm takes a few seconds to run (even for 1,000s of input genes) and requires only ca 200MiB of RAM.

tool (required)

The most important required positional argument is tool (see here for the specific versions of supported tools):

• deeparg
• argsoap
• abricate
• resfinder
• amrfinderplus
• groot
• hamronization

I/O (required)

• -i or --input: path to the annotation result
• -o or --output: the file to save normalization results

--db (optional)

This is the ARG database used to perform annotation:

• sarg
• ncbi
• resfinder
• deeparg
• megares
• argannot
• groot-core-db, groot-db, groot-resfinder, groot-argannot, groot-card

--db vs tool

ARG annotation tools can use several ARG databases for annotation. Hence, the tool is a required argument while --db is mostly optional. For some tools, however, specifiying --db is required, see below:

tool	--db
deeparg	Not required
argsoap	Not required
abricate	Any from ncbi, deeparg, resfinder, sarg, megares, argannot, or resfinderfg
resfinder	Not required
amrfinderplus	Not required
groot	Any from groot-argannot, groot-resfinder, groot-db, groot-core-db, or groot-card
hamronization	Not required

--hamronization_skip_unsupported_tool (optional)

Combined hamronization results can have ARGs detected by unsupported tools (e.g. staramr). By default, argNorm throws an exception as these are unsupported tools, however, --hamronization_skip_unsupported_tool allows users to skip rows with unsupported tools. A warning will be raised rather than an exception.

-h or --help

Use argnorm -h or argnorm --help to see available options.

>argnorm -h
usage: argnorm [-h] [--db {argannot,deeparg,megares,ncbi,resfinder,resfinderfg,sarg,groot-db,groot-core-db,groot-argannot,groot-resfinder,groot-card}] [-i INPUT] [--hamronization_skip_unsupported_tool]
               [-o OUTPUT]
               {argsoap,abricate,deeparg,resfinder,amrfinderplus,groot,hamronization}

argNorm normalizes ARG annotation results from different tools and databases to the same ontology, namely ARO (Antibiotic Resistance Ontology).

positional arguments:
  {argsoap,abricate,deeparg,resfinder,amrfinderplus,groot,hamronization}
                        tool (required): The bioinformatics tool used for ARG annotation.

options:
  -h, --help            show this help message and exit
  --db {argannot,deeparg,megares,ncbi,resfinder,resfinderfg,sarg,groot-db,groot-core-db,groot-argannot,groot-resfinder,groot-card}
                        --db (mostly optional): The database used alongside the ARG annotation tool. This is only required if abricate or groot is used as a tool. Please refer here for more information
                        on --db: https://github.com/BigDataBiology/argNorm?tab=readme-ov-file#--db-optional
  -i INPUT, --input INPUT
                        -i (required): The path to the ARG annotation result which needs to be normalized.
  --hamronization_skip_unsupported_tool
                        --hamronization_skip_unsupported_tool (optional): skip rows with unsupported tools for hamronization outputs. argNorm be default will raise an exception if unsupported tool is
                        found in hamronization. Use this if you only want argNorm to raise a warning.
  -o OUTPUT, --output OUTPUT
                        -o (required): The path to the output file where you would like to store argNorm's results

Outputs

The following columns are added to the tsv outputs of ARG annotation tools:

Table column	Description
ARO	ARO accessions of ARG
confers_resistance_to	ARO accessions of drugs to which ARGs confer resistance to
resistance_to_drug_classes	ARO accessions of drugs classes to which drugs in confers_resistance_to belong
Cut_Off	RGI cut-off scores for ARO mappings. If manually curated, cut-off score is Manual

A comment is added to the very top of the ARG annotation tool outputs specifying the argNorm version used if ran on the command line. For example:

# argNorm version: 0.6.0
input_sequence_id	input_file_name	gene_symbol	gene_name	reference_database_id ...
. REST OF ARG ANNOTATION OUTPUT TSV TABLE
.
.

NOTE: This is only if argNorm is used on the CLI, if you used argNorm's normalizers there will be no comment with the argNorm version

NOTE: THIS WILL BREAK ANY PREVIOUS SCRIPTS ANALYZING ARGNORM CLI OUTPUTS BEFORE THIS UPDATE! PLEASE USE THE skiprows=1 ARGUMENT WHEN LOADING ARGNORM OUTPUT DATAFRAMES TO IGNORE THE COMMENT WITH THE ARGNORM VERSION AS SHOWN BELOW:

import pandas as pd
df = pd.read_csv(<PATH TO ARGNORM OUTPUT>, sep='\t', skiprows=1)

Cut_Off column

• Cut-off scores are procured from RGI outputs after running input databases through RGI. These cut-off scores correspond to RGI's Discovery paradigm (learn more here: https://github.com/arpcard/rgi/blob/master/docs/rgi_main.rst).
• RGI's cut-off scores are of three types: Loose, Strict, and Perfect.
• argNorm also adds another score, Manual, to signify genes that are mapped to ARO terms manually without using RGI.

From RGI documentation (https://github.com/arpcard/rgi/blob/master/docs/rgi_main.rst):

The RGI analyzes genome or proteome sequences under a Perfect, Strict, and Loose (a.k.a. Discovery) paradigm. The Perfect algorithm is most often applied to clinical surveillance as it detects perfect matches to the curated reference sequences in CARD. In contrast, the Strict algorithm detects previously unknown variants of known AMR genes, including secondary screen for key mutations, using detection models with CARD's curated similarity cut-offs to ensure the detected variant is likely a functional AMR gene. The Loose algorithm works outside of the detection model cut-offs to provide detection of new, emergent threats and more distant homologs of AMR genes, but will also catalog homologous sequences and spurious partial matches that may not have a role in AMR. Combined with phenotypic screening, the Loose algorithm allows researchers to hone in on new AMR genes.

Within the Perfect, Strict, and Loose paradigm, RGI currently supports CARD's protein homolog models, protein variant models, protein over-expression models, and rRNA mutation models:

Protein Homolog Models (PHM) detect protein sequences based on their similarity to a curated reference sequence, using curated BLASTP bitscore cut-offs, for example NDM-1. Protein Homolog Models apply to all genes that confer resistance through their presence in an organism, such as the presence of a beta-lactamase gene on a plasmid. PHMs include a reference sequence and a bitscore cut-off for detection using BLASTP. A Perfect RGI match is 100% identical to the reference protein sequence along its entire length, a Strict RGI match is not identical but the bit-score of the matched sequence is greater than the curated BLASTP bit-score cutoff, Loose RGI matches have a bit-score less than the curated BLASTP bit-score cut-off.

Example 1: argNorm as a command line tool

Here is a quick demo of running argNorm on the command line.

Step 1: Install argNorm

Install argNorm and check installation

pip install argnorm
argnorm -h

argnorm -h or argnorm --help will display all the available options to run argNorm with.

> argnorm -h
usage: argnorm [-h] [--db {argannot,deeparg,megares,ncbi,resfinder,resfinderfg,sarg,groot-db,groot-core-db,groot-argannot,groot-resfinder,groot-card}] [-i INPUT] [--hamronization_skip_unsupported_tool]
               [-o OUTPUT]
               {argsoap,abricate,deeparg,resfinder,amrfinderplus,groot,hamronization}

argNorm normalizes ARG annotation results from different tools and databases to the same ontology, namely ARO (Antibiotic Resistance Ontology).

positional arguments:
  {argsoap,abricate,deeparg,resfinder,amrfinderplus,groot,hamronization}
                        tool (required): The bioinformatics tool used for ARG annotation.

options:
  -h, --help            show this help message and exit
  --db {argannot,deeparg,megares,ncbi,resfinder,resfinderfg,sarg,groot-db,groot-core-db,groot-argannot,groot-resfinder,groot-card}
                        --db (mostly optional): The database used alongside the ARG annotation tool. This is only required if abricate or groot is used as a tool. Please refer here for more information
                        on --db: https://github.com/BigDataBiology/argNorm?tab=readme-ov-file#--db-optional
  -i INPUT, --input INPUT
                        -i (required): The path to the ARG annotation result which needs to be normalized.
  --hamronization_skip_unsupported_tool
                        --hamronization_skip_unsupported_tool (optional): skip rows with unsupported tools for hamronization outputs. argNorm be default will raise an exception if unsupported tool is
                        found in hamronization. Use this if you only want argNorm to raise a warning.
  -o OUTPUT, --output OUTPUT
                        -o (required): The path to the output file where you would like to store argNorm's results

Step 2: Create working directory & download sample data

argNorm adds ARO mappings and drug categories as additional columns to the outputs of ARG annotation tools.

For this example, we will run argNorm on the ARG annotation output from the ResFinder tool (with the ResFinder database).

Create a folder called argNorm_tutorial and store the downloaded data file in it. Navigate into the argNorm_tutorial folder.

mkdir argNorm_tutorial
cd argNorm_tutorial

Click here to download the input data.

If you are on Linux:

wget https://raw.githubusercontent.com/BigDataBiology/argNorm/main/examples/raw/resfinder.resfinder.orfs.tsv

Step 3: Running argNorm

Here is a basic outline of most argNorm commands:

argnorm [tool] -i [original_annotation.tsv] -o [argnorm_result.tsv] [--db]

Here, tool refers to the ARG annotation tool used (ResFinder in this case). original_annotation.tsv is the path to the input data and argnorm_result.tsv is the path to output file where the resulting table from argNorm will be stored. --db is the ARG databases used along with tool to perform annotation. ResFinder does not require a --db (argNorm will automatically load up the ResFinder database), however, --db is required for the ARG annotation tools groot and abricate.

To run argNorm on the input data, use this command in your terminal:

argnorm resfinder -i ./resfinder.resfinder.orfs.tsv -o ./resfinder.resfinder.orfs.normed.tsv

The argNorm result will be stored in the file resfinder.resfinder.orf.normed.tsv.

Example 2: argNorm as a Python library

Code

Save this piece of Python code to a file called argnorm_tutorial.py

# Import from argNorm
from argnorm.lib import map_to_aro
from argnorm.drug_categorization import confers_resistance_to, drugs_to_drug_classes

# Creating a list of input genes to be mapped to the ARO
list_of_input_genes = ['sul1_2_U12338', 'sul1_3_EU855787', 'sul2_1_AF542061']

# The database from which the `list_of_input_genes` was created is the ResFinder database
database = 'resfinder'

# Looping through `list_of_input` genes and mapping each gene to the ARO
# Storing each ARO mapping in the `list_of_aros` list
list_of_aros = []
for gene in list_of_input_genes:
    # Using `id` attribute to get ARO number
    # `name` attribute can be used to get name of gene in ARO
    list_of_aros.append(map_to_aro(gene, database).id)
print(list_of_aros)

# Looping through `list_of_aros` and finding the drugs to which the each ARO confers resistance to
# Storing each drug in the `list_of_drugs` list
list_of_drugs = []
for aro in list_of_aros:
    list_of_drugs.append(confers_resistance_to(aro))
print(list_of_drugs)

# Looping through `list_of_drugs` and finding the superclass/drug class of each drug
# Storing each superclass/drug class in the `list_of_drug_classes` list
list_of_drug_classes = []
for drug in list_of_drugs:
    list_of_drug_classes.append(drugs_to_drug_classes(drug))
print(list_of_drug_classes)

Explanation

To use argNorm as a library, we must first import it in our Python file:

from argnorm.lib import map_to_aro
from argnorm.drug_categorization import confers_resistance_to, drugs_to_drug_classes

The lib module contains the function map_to_aro which will return the ARO number of a particular antibiotic resistance gene. The drug_categorization module contains the functions confers_resistance_to and drugs_to_drug_classes. The confers_resistance_to function returns the drugs to which a gene confers resistance to. The drugs_to_drug_classes function returns the drug class to which a specific drug belongs.

The map_to_aro function takes two arguments: gene and database. gene is the name of an antibiotic resistance gene. database is the database from which gene is taken from.

In this example, a list of genes from the ResFinder database is used, and map_to_aro maps each gene in the list to an ARO term.

The ARO numbers from the ARO terms are also stored in the list_of_aros list. The ARO numbers can be accessed using the id attribute of the ARO terms. The names of the gene of the ARO terms an be accessed using the name attribute.

database = 'resfinder'

list_of_aros = []
for gene in list_of_input_genes:
    list_of_aros.append(map_to_aro(gene, database).id)
print(list_of_aros)

Once a list of ARO numbers is created for each gene, the confers_resistance_to function can be used on each ARO number to create a list of drugs to which each gene/ARO confers resistance to:

list_of_drugs = []
for aro in list_of_aros:
    list_of_drugs.append(confers_resistance_to(aro))
print(list_of_drugs)

Now, each drug in the list_of_drugs can be categorized into a broader drug category using the drugs_to_drug_classes function:

list_of_drug_classes = []
for drug in list_of_drugs:
    list_of_drug_classes.append(drugs_to_drug_classes(drug))
print(list_of_drug_classes)

Output

List of ARO numbers
['ARO:3000410', 'ARO:3000410', 'ARO:3000412']

Confers resistance to
[['ARO:3000324', 'ARO:3000325', 'ARO:3000327', 'ARO:3000329', 'ARO:3000330', 'ARO:3000683', 'ARO:3000684', 'ARO:3000698', 'ARO:3000699'], ['ARO:3000324', 'ARO:3000325', 'ARO:3000327', 'ARO:3000329', 'ARO:3000330', 'ARO:3000683', 'ARO:3000684', 'ARO:3000698', 'ARO:3000699'], ['ARO:3000324', 'ARO:3000325', 'ARO:3000327', 'ARO:3000329', 'ARO:3000330', 'ARO:3000683', 'ARO:3000684', 'ARO:3000698', 'ARO:3000699']]

Resistance to drug classes
[['ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282'], ['ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282'], ['ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282', 'ARO:3000282']]

API Reference

Click here to view the API Reference

Authors

• Vedanth Ramji [e-mail]*
• Hui Chong [e-mail]
• Svetlana Ugarcina Perovic [e-mail]
• Finlay Maguire [e-mail]
• Luis Pedro Coelho [e-mail]

*: current maintainer