FUNGAR: a pipeline for detecting antifungal resistance mutations directly from metagenomic short reads

Motivation: Antifungal resistance has become an increasing global concern in both clinical and environmental health. Detecting known resistance mutations directly from sequencing reads, in special metagenomic samples, remains a major challenge. As fungal pathogens are often neglected compared with bacterial pathogens, most available tools are designed for bacterial taxa, whereas tools targeting fungi typically require assembled genomes. In metagenomic datasets, assembly-based strategies may result in substantial information loss due to genome fragmentation, low-abundance species, or incomplete recovery of resistance loci. Results: Here, we present FUNGAR, an open-source pipeline for the rapid identification of antifungal resistance genes and mutations directly from short-read data. FUNGAR employs translated alignments with DIAMOND and curated data from the FungAMR database to detect amino acid substitutions across all six open reading frames. The pipeline produces structured, reproducible reports linking detected variants to their associated antifungal drugs and can be easily customized for new species or databases.

💡 Research Summary

Antifungal resistance is an escalating threat to both human health and the environment, yet tools for detecting resistance mutations directly from metagenomic short‑read data are scarce. Most existing pipelines are either bacterial‑centric or require assembled fungal genomes, which leads to loss of information when dealing with low‑abundance or fragmented genomes. To address this gap, the authors present FUNGAR, an open‑source Bash/Python pipeline that identifies known antifungal resistance genes (AfRGs) and their amino‑acid substitutions directly from raw FASTQ files.

The core of FUNGAR is a translated alignment step performed with DIAMOND in blastx mode. Reads are translated in all six possible open reading frames (ORFs) and aligned against a protein database derived from the curated FungAMR repository. By default, only alignments covering at least 50 amino acids (appropriate for 150‑bp reads) are retained, but users can adjust minimum query coverage, percent identity, and the genetic code used for translation. After alignment, a Python script parses the DIAMOND output, cross‑references each hit with the mutation list in FungAMR (gene, position, reference amino‑acid, variant, associated drug), and produces two CSV reports: a detailed per‑read table and a summary table aggregating read counts per mutation. The pipeline ships with pre‑built DIAMOND databases for common clinical and environmental fungi, yet it also accepts user‑provided databases (a .dmnd protein file plus a CSV mutation list), allowing rapid adaptation to emerging species or novel resistance mechanisms.

Validation began with a synthetic dataset. The authors engineered eight paired‑end 150‑bp reads covering the Pneumocystis jirovecii dihydrofolate reductase (DHFR) gene, inserting two mutations at the second residue: D2E (known to confer methotrexate resistance) and D2F (a neutral change). All reads were given a Phred quality of 30. FUNGAR correctly identified only the D2E mutation, demonstrating both specificity and the ability to detect variants regardless of strand orientation or ORF.

Real‑world performance was assessed on two independent collections. Dataset G comprised ten Aspergillus fumigatus isolates, eight of which were phenotypically azole‑resistant. Using the A. fumigatus‑specific database, FUNGAR recovered the canonical Cyp51A L98H mutation and additionally reported resistance‑associated variants in Fks, Hmg1, Hmg2, and Sdh that were not highlighted in the original study. Notably, it also flagged mutations in samples previously classified as wild‑type, suggesting hidden resistance potential.

Dataset M consisted of metagenomic samples from three cystic fibrosis patients (11 samples total) colonized primarily by Aspergillus and Candida species. Separate databases for each genus were built. FUNGAR detected β‑tubulin F200Y in patient A (two samples), Sdh H270R in patients B and C, and Candida Erg1 D153E in patient C, all of which are linked to azole resistance. These findings illustrate the pipeline’s capacity to parse mixed‑species metagenomes and to provide clinically relevant resistance profiles without prior genome assembly.

In conclusion, FUNGAR is the first pipeline that directly calls known antifungal resistance mutations from short‑read metagenomic data. Its reliance on translated DIAMOND alignments ensures speed and sensitivity, while the modular database design guarantees extensibility. Limitations include a focus on protein‑level point mutations (excluding copy‑number changes, promoter alterations, or epigenetic mechanisms) and dependence on the completeness of the underlying FungAMR catalog. Nonetheless, the tool offers a rapid, reproducible, and customizable framework for surveillance of emerging antifungal resistance in both environmental and clinical contexts. The source code, documentation, and pre‑built databases are freely available at https://github.com/resgen-br/fungar, and the authors encourage community contributions to keep the resource up‑to‑date.