MetaPointFinder

MetaPointFinder is a tool for detecting and scoring resistance-associated point mutations directly from long-read and short-read metagenomics sequencing data, using the AMRFinder database as reference.

A user-friendly web interface is available at klif.uu.nl/metapointfinder.

---

Features

• Detects amino acid substitutions in translated reads using DIAMOND and AMRFinder protein mutation databases.
• Detects nucleotide mutations using KMA against AMRFinder DNA databases.
• Scores mutations using MSA-based alignment and outputs both read-level classifications and summary tables.
• Supports direct analysis of metagenomic datasets (nanopore or Illumina fastq files).

---

Installation

Conda/mamba

The simplest installation methods are Mamba and Docker. These are therefore recommended. Both methods include MetaPointFinder itself along with all its dependencies).

To install Mamba, see its documentation.

To install MetaPointFinder using mamba, type:

mamba create -n metapointfinder -c conda-forge -c bioconda metapointfinder

or with conda

conda create -n metapointfinder -c conda-forge -c bioconda metapointfinder

Docker

To install Docker, see its documentation. The Docker container of MetaPointFinder can be acquired through GitHub:

docker pull ghcr.io/aldertzomer/metapointfinder:main

(Also see the package page .)

Dependencies

• R with libraries:

  • Biostrings
  • pwalign
  • parallel

• DIAMOND

• KMA

• wget

For your convenience, we have compiled a conda file that can install all dependencies in one go:

git clone https://github.com/aldertzomer/metapointfinder.git
cd metapointfinder
mamba env create -f dependencies.yaml

This practically installs MetaPointFinder for you. You can then run it as follows:

mamba activate metapointfinder
python3 metapointfinder.py --help

(However, this will only work when 'metapointfinder' is your current working directory. It is not automatically linked to your environment ($PATH).)

R packages can be installed separately with:

install.packages("parallel")
install.packages("BiocManager")
BiocManager::install(c("pwalign", "Biostrings"))

DIAMOND can be downloaded from: https://github.com/bbuchfink/diamond/releases/

KMA can be obtained from: https://bitbucket.org/genomicepidemiology/kma

Operating system compatibility

MetaPointFinder has been tested on Linux only, specifically on:

• Red Hat Enterprise Linux (RHEL) version 9.5
• Fedora version 43
• Alpine Linux v3.22 (as Docker container)

It is expected to work on any GNU/Linux system and possibly other nix-like systems. The Docker container should work on any platform.

---

Usage

metapointfinder.py --input file.fastq[.gz] --db databasefolder --output outputfolder --identity 85 --threads 4 [--force]

When using the Docker container, you can use these commands to get started:

First, check the ID of the image, for example:

docker images

Then run the image that is listed as ghcr.io/aldertzomer/metapointfinder like so, where you fill in the ID that was returned by the previous command:

docker run [fill in your ID here!] python3 metapointfinder.py

Finally, to run the actual program using FASTQ files in your current directory, run:

docker run -v $(pwd):/workdir:z [fill in your ID here!] python3 metapointfinder.py\
 --input /workdir/[example].fastq.gz --output /workdir/output\
 --db /workdir/metapointfinder_database --identity 85 --threads 4

Make sure to fill in the right paths to your FASTQ file, desired output directory and database folder!

Arguments

• --help: Print the usage
• --version: Shows the program version
• file.fastq or file.fastq.gz: sequence read data file. Currently paired read data information is not used.
• databasefolder: Directory where AMRFinder databases will be downloaded/prepared.
• output_folder: Directory to store all results.
• --identity 85: percentage identity cutoff. Results may vary with different cutoffs
• --threads 4 : parallel processing in diamond, kma and calling pwalign in parallel using mcapply in R
• --force: write in same directory and overwrite results. Use with care. Never use in parallel with different input files.

If databasefolder does not exist, it will be created automatically and the required AMRFinder databases will be downloaded and processed.

---

Pipeline Overview

1. Database Preparation:

   • AMRFinder protein and DNA reference databases are downloaded.
   • Mutation tables are preprocessed for scoring.
   • DIAMOND and KMA databases are created.

2. Read Processing:

   • Reads aligned to AMR proteins using DIAMOND blastx.
   • Reads aligned to AMR genes using KMA.
   • Matching regions extracted and scored using the provided R scripts.

3. Mutation Scoring (in R):

   • Translated protein reads and DNA reads are aligned to references using ClustalW (via the pwalign R package).
   • Known mutations from AMRFinder mutation tables are searched for.
   • Each read is scored:
     • MutationScore: Number of known resistance mutations detected.
     • DetectedMutations: List of mutations found (or "None").
     • WTConfirmedPositions: Number of WT positions detected.
     • Status: Scoring of read as Wildtype (WT), Resistant (R) or Unknown (UNKNOWN)

4. Results Aggregation:

   • WT, R and UNKNOWN counts summarized by antibiotic class and gene.

---

Output Files

File	Description
*.prot.updated_table_with_scores_and_mutations.tsv	Read-level protein mutation classification (WT/R/UNKNOWN).
*.dna.updated_table_with_scores_and_mutations.tsv	Read-level DNA mutation classification (WT/R/UNKNOWN).
*.class.prot.summary.txt	WT/R summary per antibiotic class (protein level).
*.gene.prot.summary.txt	WT/R summary per gene (protein level).
*.class.dna.summary.txt	WT/R summary per antibiotic class (DNA level).
*.gene.dna.summary.txt	WT/R summary per gene (DNA level).
Log and error files	Detailed logs for troubleshooting DIAMOND, KMA, and R execution.

---

Example Output (Class Summary)

Example (*.class.prot.summary.txt):

class           WT    R     UNKNOWN
BETA-LACTAM     255   44    2003
COLISTIN        59    12    2938
QUINOLONE       185   25    3209
MULTIDRUG       134   33    2635

---

Example Output (Read Classification)

Example (*.prot.updated_table_with_scores_and_mutations.tsv):

class	gene	read	reference	target	changes_str	MutationScore	DetectedMutations	WTConfirmedPositions	Status
BETA-LACTAM	two-component_system_sensor_histidine_kinase_BaeS	10478f68-f181-4626-976f-93d14c49844b	...	...	Y42H,T175P,R153P	0	None	3	WT
COLISTIN	two-component_system_sensor_histidine_kinase_PmrB	1249bd62-efe5-46d5-96c4-1e903c85dec5	...	...	V161G,T92P	1	V161G	2	R
...	...	...	...	...	...	...	...	...

• class: Antibiotic class.
• gene: Gene or protein name.
• read: Read identifier.
• reference: Reference sequence used for alignment.
• target: Sequence from read.
• changes_str: List of expected mutations.
• MutationScore: Number of detected resistance mutations.
• DetectedMutations: Mutations actually found in the read.
• WTConfirmedPositions: Number of WT positions detected.
• Status: Scoring of read as Wildtype (WT), Resistant (R) or Unknown (UNKNOWN)

---

Notes

• Databases are downloaded directly from NCBI during the first run.
• Mutation detection relies on known mutations in the AMRFinder database. Novel mutations will not be scored.
• For metagenomic samples with highly fragmented reads, alignment quality may affect detection sensitivity.
• There is a correlation between fragmentation and number of reads scored as WT and UNKNOWN.

---

License

This tool is released under the Apache-2.0 License.

---

Citation

If you use MetaPointFinder in your work, please cite:

Zomer, A.L. et al. MetaPointFinder: a tool to detect resistance mutations directly from long-read metagenomic data using AMRFinder databases. GitHub Repository