Quantitative comparison of fungal genome assembly strategies using short and long-reads from simulated and empirical sequencing data
High-quality fungal reference genomes are essential for comparative, functional, and evolutionary studies, yet fungal genome features such as repeats, structural rearrangements, accessory chromosomes, and intron-rich genes can complicate genome assembly and the selection of cost-effective sequencing strategies. Here, we benchmark fungal genome assembly performance using simulated and empirical short- and long-read datasets to evaluate how sequencing depth, assembler choice, and genome characteristics influence contiguity, completeness, accuracy, and computational requirements. Using simulated reads from complete fungal genomes spanning diverse sizes and compositions, we evaluated short-read, long-read, hybrid, and polished long-read assemblies across sequencing depths from 10X to 100X. Key trends were validated using empirical sequencing data from 10 fungal isolates assembled with multiple strategies, including different Flye assembler parameter sensitivity and short-read polishing. Across datasets, long reads produced the largest improvements in contiguity, with most gains achieved at [~]20-40X coverage and diminishing returns beyond moderate depth. Short-read polishing substantially improved base-level accuracy at relatively low cost, with [~]10-20X coverage often sufficient to approach maximal error reduction. Hybrid assemblers showed strong algorithmic variability, with trade-offs between contiguity, error rates, and computational demand. Genome architecture also influenced outcomes, as larger and more feature-dense genomes benefited more from long-read data while GC content had limited impact. Overall, our results suggest that moderate long-read coverage ([~]30-40X) combined with modest short-read polishing ([~]10-20X), particularly using Flye plus Polypolish, provides a strong balance of contiguity, completeness, accuracy, and resource efficiency for generating high-quality fungal genome assemblies. Impact statementFungal genome sequencing is expanding rapidly across ecology, plant pathology, biotechnology, and clinical and veterinary microbiology, yet experimental design decisions regarding sequencing depth, assembler selection, and hybrid workflows are still largely guided by bacterial benchmarking studies or limited single-species comparisons. Because fungal genomes vary widely in size, repeat content, and gene architecture, these assumptions can lead to inefficient sequencing strategies, increased computational costs, and suboptimal assemblies. Here, we develop a reproducible assembly benchmarking framework that combines large-scale simulations from 66 complete fungal genomes spanning plant, animal, and human-associated taxa with newly generated short- and long-read sequencing data from 10 field-collected isolates. This approach enables evaluation of assembler performance across diverse genome architectures and tests whether patterns identified in simulations translate to real biological datasets. Across both simulated and empirical datasets, we show that reliable fungal genome reconstruction can be achieved without excessive sequencing depth by identifying consistent performance thresholds. Assembly contiguity and completeness stabilize at moderate long-read coverage, after which improvements depend more strongly on assembler choice and genome structure than on additional data volume. Hybrid workflows show trade-offs in accuracy, contiguity, and computational demand, whereas targeted short-read polishing provides an efficient strategy for improving base-level accuracy. These findings offer practical guidance for fungal genome assembly and support more robust downstream genomic analyses across non-model microbial systems, including ecologically, agriculturally, and clinically important fungi. Data summaryThe reference fungal genomes used for simulation are available in the NCBI Assembly database under the accession numbers listed in Supplementary Table S1. Empirical raw sequencing data generated for this study are deposited in the NCBI Sequence Read Archive (SRA) under accession PRJNA1474061. All scripts used for read simulation, assembly, polishing, benchmarking, and statistical analyses are available in the project GitHub repository (github.com/bielasilva/fungi_assembly_benchmarking). Software versions, parameters, and workflow configurations are provided within the repository and detailed in the Methods.