New pipeline detects antifungal resistance mutations directly from metagenomic sequencing reads

In the escalating battle against drug-resistant fungi, one of the most stubborn obstacles has been simply seeing the enemy clearly. Bacterial antimicrobial resistance can be screened with a growing arsenal of bioinformatic tools, but fungi have lagged far behind: most existing software assumes the analyst is working with a fully assembled genome, a luxury that rarely exists in the messy, fragmented world of metagenomic data. A team of Brazilian researchers at the Federal University of Rio Grande do Sul has now built a solution designed specifically for this gap. The tool, called FUNGAR, is an open-source pipeline that can detect known antifungal resistance genes and mutations directly from short-read sequencing data, without requiring genome assembly, and it is described in a peer-reviewed article in the journal BMC Bioinformatics.

The clinical and environmental stakes of antifungal resistance are considerable and rising. Resistance mutations in fungal pathogens undermine the treatment of invasive infections, particularly in immunocompromised patients, and there is growing concern about the spread of resistance determinants between agricultural and clinical settings, since several classes of antifungal compounds are used in both worlds. Surveillance efforts that might track these mutations in hospitals, soils, water systems, or wastewater currently depend on laborious culture-based methods or on assembly-dependent genomic workflows that lose substantial information when species are rare in a sample or when resistance loci fail to assemble completely. FUNGAR was conceived to bypass that bottleneck entirely and to allow resistance screening to run straight from the raw sequencing reads that metagenomic studies already generate in enormous quantities.

The technical heart of the pipeline is a strategy of translated alignment. Rather than trying to align nucleotide reads against gene sequences and infer protein-level changes afterward, FUNGAR translates sequencing reads into all six possible open reading frames, the three forward frames and three reverse frames of any stretch of DNA, and then aligns the resulting amino acid sequences against curated protein references using DIAMOND, a fast sequence aligner optimized for large-scale protein comparisons. The curated reference data come from the FungAMR database, a specialized resource of antifungal resistance determinants. Working at the protein level matters because the biologically meaningful events, amino acid substitutions that confer resistance, are precisely what the tool is designed to call. Reading all six frames means that the pipeline does not need to know in advance which strand or reading frame a resistance gene occupies, a significant advantage when the provenance of reads in a metagenomic mixture is unknown.

Because every high-throughput detector must balance sensitivity against false alarms, the developers built in a configurable read-support threshold, the minimum number of reads that must back a putative mutation before it is reported. Based on their benchmarking, the team recommends a threshold of at least three supporting reads for metagenomic datasets, a setting that suppresses spurious calls arising from sequencing errors while still retaining genuine low-frequency variants. In addition, FUNGAR applies paired-end mate-concordance filtering. Modern sequencing platforms typically produce reads in pairs that originate from the same DNA fragment at a predictable distance and orientation; requiring that both members of a pair support the same variant provides a strong, computationally cheap check against artifacts. Together these filters substantially reduce the false-positive rate, which the authors quantify using a companion benchmarking script that stochastically generates synthetic reads across a range of sequencing depths and evaluates precision and false positives at each depth.

The pipeline also addresses a dimension of antifungal resistance that is often overlooked in automated tools: context. A mutation that matters clinically, meaning it alters susceptibility to drugs used in human medicine, may sit in the same gene family as mutations relevant only to agricultural fungicides. FUNGAR automatically classifies each detected variant according to its drug-class context, distinguishing clinical from agricultural associations, and links variants to their associated antifungal compounds. This automatic annotation turns a list of raw substitutions into something closer to an actionable surveillance report, telling an epidemiologist not merely that a mutation exists but which therapeutic or agrochemical classes it potentially threatens. Results are compiled into a self-contained HTML report, making the output readable and shareable without additional software.

Reproducibility was another design priority. The software is written in Bash and Python 3, is distributed under the permissive MIT license, and can be installed through the Conda and Bioconda package managers, with DIAMOND as its principal external dependency. It was built and tested on Ubuntu 24.04 LTS. The authors note that tools such as fastp, Bowtie2, and Kraken2, which they used for read pre-processing, host depletion, and taxonomic characterization of the datasets analyzed in the study, are not invoked by FUNGAR itself, keeping the pipeline lean and composable with whatever upstream workflows a laboratory already runs. For teams generating metagenomic data on clinical isolates, environmental samples, or wastewater, integrating resistance screening into an existing pipeline becomes a matter of appending a single tool rather than reengineering the workflow.

The validation work demonstrates the pipeline’s range across several data types. In one benchmark, the authors constructed synthetic FASTQ files manually seeded with mutations in the dihydrofolate reductase protein of Pneumocystis jirovecii, an opportunistic fungal pathogen of significant concern, and confirmed that FUNGAR recovered the planted variants. In a genomic test, the team screened previously published data from Aspergillus fumigatus, the mold responsible for the devastating infection aspergillosis and a notorious case study in the agricultural-clinical crossover of azole resistance. Finally, the pipeline was applied to real environmental metagenomic data and to clinical metagenomic data, providing proof that the approach functions not only on curated benchmarks but on the kind of heterogeneous, multi-species material that surveillance programs actually produce.

Statistical rigor informed the benchmarking as well. The authors used analysis of variance with honestly significant difference post hoc testing to compare performance across conditions, and they tracked both precision and false-positive rate as functions of sequencing depth. This depth-dependence is a crucial consideration in metagenomics, where coverage of any given resistance gene can vary by orders of magnitude depending on the organism’s abundance in the sample. By characterizing where the read-support threshold reliably separates true signal from noise, the study gives practitioners a defensible basis for choosing parameters rather than relying on intuition. The stochastic read-generation approach to benchmarking also means the evaluation can be repeated and extended by other groups as sequencing technologies evolve.

According to the authors, FUNGAR is, to their knowledge, the first pipeline capable of detecting and annotating known mutations in antifungal resistance genes directly from short-read sequencing data. That distinction matters because it removes the assembly step that has historically been the weak link for fungal metagenomics. Genome assembly from short reads is inherently lossy: low-abundance species contribute too few reads to assemble contiguously, repetitive regions fragment into unresolved pieces, and resistance loci can be scattered across fragments too short to identify. By working at the read level with translated alignments, FUNGAR sidesteps these failure modes, and the authors argue that the result is a fast, reproducible, and extensible framework suited to monitoring emerging antifungal resistance mechanisms in both genomic and metagenomic samples.

The broader implications reach into public health preparedness at a moment when fungal threats are commanding unprecedented attention. Resistance mechanisms in fungi such as Aspergillus fumigatus, Candida species, and Cryptococcus are increasingly recognized as a One Health problem, linking agricultural fungicide use, environmental reservoirs, and clinical treatment failure into a single interconnected system. Tools that can cheaply and rapidly survey metagenomes from soil, air, water, and hospitals for known resistance mutations give that emerging surveillance architecture a practical foundation. The work was supported by Brazil’s National Council for Scientific and Technological Development and by the National Institute of Science and Technology in Human Pathogenic Fungi as part of a One Health antimicrobial resistance genomic surveillance initiative, underscoring the institutional commitment behind the effort. The software is freely available, the underlying article is open access under a Creative Commons license, and the authors, Henrique RM Antoniolli, Lívia Kmetzsch, and Charley C Staats, report no competing interests. For researchers confronting the quiet spread of antifungal resistance, FUNGAR offers something they have long lacked: a way to interrogate raw sequencing reads for fungal resistance mutations directly, quickly, and at scale.