Pangenomic Graph Softwares as Passport for Shared Global Applicability

1. Introduction

For decades, plant genomics has been organized around a simple concept: identify a representative reference genome and map genetic variation against it. This strategy has generated enormous advances in plant biology and crop improvement. However, the assumption that one genome can adequately represent a species is increasingly difficult to defend owing to the diversity across plant genomes. Variation extends far beyond single-nucleotide polymorphisms (SNPs) to include insertions, deletions, duplications, inversions, copy-number changes, transposable elements, presence/absence variation and highly divergent haplotypes (Cochetel et al. 2026). Some of these variants can involve genes or regulatory regions that are completely absent from the commonly used reference genome. Consequently, a reference-centered strategy can systematically underestimate the genetic diversity available within a crop species (Kaur et al.2024). Plant pangenomics emerged as a response to this limitation. Rather than representing a species by one genome, a pangenome attempts to capture genomic diversity across multiple individuals. Plant pangenomics has become an essential framework for capturing genetic diversity beyond a single reference genome, including presence–absence variation (PAV), structural variation (SV), copy-number variation, repetitive sequences, and divergent genomic regions. Advances in long-read sequencing and chromosome-scale assembly have enabled high-quality pangenomes and increasingly sophisticated graph-based representations, supporting studies of adaptation, domestication, disease resistance, and agronomic traits (Kaur et al. 2024, Cochetel et al. 2026) (Figure 1 and Figure 2). Recent developments in long-read sequencing and genome assembly have made it possible to construct increasingly complete and haplotype-resolved genome resources. Graph-based approaches have further expanded this concept by allowing alternative genomic sequences to be represented within a unified structure. Current approaches include reference-guided graph construction using Minigraph and Minigraph-Cactus (Hickey et al. 2024), and reference-unbiased construction using PGGB. Minigraph is computationally efficient and well suited for large-scale detection of structural and presence–absence variation, but its reference dependence can introduce bias and its original implementation has limited resolution for small variants. Minigraph-Cactus combines efficient structural-variant graph construction with base-level whole-genome alignment, enabling representation of SNPs, small indels, and large SVs and therefore providing a strong balance between resolution and scalability. PGGB (Garrison et al. 2024), based on all-versus-all whole-genome alignment, minimizes reference bias and is particularly valuable for highly divergent genomes, although it requires substantially greater computational and memory resources. Downstream tools such as VG (Garrison et al. 2018) and ODGI (Guarracino et al. 2022) support graph analysis, while SyRI (Goel et al. 2019) and SVIM-asm (Heller and Vingron 2020) can provide independent SV detection and validation. Overall, plant pangenomics is shifting from genomic cataloguing toward pangenome-assisted breeding, enabling discovery and deployment of favourable alleles and haplotypes through integration with transcriptomic, phenotypic, and other multi-omics data. Major challenges remain for large, repetitive, heterozygous, and polyploid genomes, particularly in graph scalability, standardization, storage, and reproducible analysis.

In our opinion, plant pangenomics is now approaching an important turning point. The central question should no longer be simply “How can we construct a larger pangenome?” Instead, it should be “How can we convert pangenomic diversity into biological knowledge and breeding value?” Artificial intelligence (AI) and machine learning are increasingly being incorporated into genomic prediction and biological interpretation. Pangenomes provide potentially valuable input data for these approaches because they capture a broader range of genetic variation. AI could help prioritize candidate variants, predict functional consequences, identify genotype–phenotype relationships and improve genomic selection models (Kaur et al. 2024). Nevertheless, we argue against treating AI as a solution to the fundamental challenges of pangenomics.

Artificial intelligence and machine learning may further support variant prioritization, functional prediction, genotype–phenotype modeling, and genomic selection, but their effectiveness will depend on high-quality, representative genomic and phenotypic datasets and should complement rather than replace biological validation. Major challenges remain, particularly the computational complexity of large, repetitive, and polyploid plant genomes, graph scalability, functional validation of novel genes and SVs, inconsistent assembly and annotation pipelines, and the lack of standardized approaches for pangenome construction, variant representation, nomenclature, and phenotype metadata. Thus, pangenome quality should be evaluated by diversity, assembly quality, functional information, and phenotype integration rather than genome number alone (Mun et al. 2022). Ultimately, the next generation of plant pangenomics should establish a pan-genome-to-phenome infrastructure that connects diverse germplasm and haplotype-resolved genomes to graph-based variation, multi-omics, phenotyping, causal variant discovery, genomic prediction, genome editing, and precision breeding, transforming the pangenome from a catalogue of genomic diversity into a predictive biological and breeding platform. In this paper, we present state of the pangenome detection softwares, their underlying implementation, application, scalibility and introduces panscape, which along with the two bash based pipelines become a complete pangenome detection kit for the plant genomes. For pangenome comparison and structural analysis, panscape can summarize and compare PAF alignments, extract aligned regions, calculate harmonic-mean estimates, and convert PAF alignments into network formats. It also analyzes GFA pangenome graphs by reporting segment, link, arc, rank, and total sequence-length statistics. Additional functionality includes generation of pangenome BED files, ancestral-state analysis, VCF comparison, and construction of a local SQLite-based pangenome read database.

Pangenomics softwares: State of art of pangenome detection algorithms

Pangenomics aims to characterize the entire set of genes and sequence variations within a species or population, moving beyond the limitations of single reference genomes which often suffer from reference bias (Bao and Weigel 2025). The field has seen a rapid expansion of software tools designed to handle the increasing volume of high-quality assemblies and long-read sequencing data (Bao and Weigel 2025, Andreace et al. 2023). Graph-based pangenomes, which represent genomic variation as nodes and edges, have become the primary focus of development, allowing for more accurate mapping and variant calling, particularly for structural variants (Hickey et al. 2019, Eizenga et al. 2020).

Pangenome software has been well established from graph construction to remapping and downstream analysis. Graph pangenomes represent an important technological development because they provide a framework for representing alternative genomic sequences. Unlike a linear reference, a graph can contain multiple allelic paths and divergent genomic regions. This makes graph-based approaches particularly attractive for species with substantial structural variation. Among graph construction, variation graphs, de Bruijn graphs and genome-graph frameworks have been widely exploited for pangenome construction. The VG toolkit is a versatile framework for constructing variation graphs from VCF files or assembly alignments (Hickey et al. 2019). Alognside, Minigraph utilizes approximate mapping to a backbone genome to add large structural variants (>50 bp) (Bao and Weigel 2025, Andreace et al. 2023). Minigraph-Cactus (Hickey et al. 2024) and PGGB (Pangenome Graph Builder) employ base-level alignments to create more comprehensive representations that include small variants and precise breakpoints (Bao and Weigel 2025, Andreace et al. 2023, Crysnanto et al. 2022). Minigraph-Cactus sits between the two, combining minigraph’s fast structural-variant skeleton with Cactus’s base-level alignment refinement, and it has become a particularly strong fit for plants specifically because of how it treats repetitive sequence. This “base-graph” strategy represents structural variations as alternate paths and allows multi-layered alignment of nested transposable elements that would otherwise be collapsed or misaligned in linear reference models, and by leveraging these graph-aware algorithms researchers can precisely quantify presence/absence variation frequency and identify introgression events across diverse cultivars, turning the plant pangenome from a conceptual inventory into a functional tool for marker-assisted selection. This TE-aware handling matters a lot for crops like sorghum, where repeat content constitutes roughly 61% of the genome and significantly influences structural variant landscapes .(Figure 2). Tools like Bifrost and Cuttlefish focus on constructing compacted and colored DBGs, which are highly efficient for storage and querying of large genomic datasets (Outten and Wareen 2021, Andreace et al. 2023). mdbg uses minimizer-space DBGs to achieve extremely fast construction and low memory usage, though it may discard small variants (Andreace et al. 2023). pangene identifies gene-level variations like orientation and copy-number changes by constructing gene graphs from protein sequences (Li et al. 2024).

One of the strongest arguments for pangenomics is that biologically important variation is not restricted to SNPs. Structural variants can alter gene dosage, gene structure, regulatory regions and genome organization. Presence/absence variation can also determine whether an individual carries entire genes or genomic regions that are missing from another genotype. Despite this importance, many conventional breeding pipelines remain strongly SNP-centric. In our view, this represents a major mismatch between the complexity of plant genomes and the tools routinely used for crop improvement.Efficiently mapping reads to a pangenome graph and calling variants are critical tasks for population genomics. VG Giraffe is a state-of-the-art mapper for human-scale pangenomes, achieving speeds comparable to linear mappers while improving variant calling (Chang et al. 2025). GraphAligner excels in long-read sequence-to-graph alignment using a minimizer-based seeding strategy (Chang et al.2025, Outten and Warren 2021). HISAT2 employs a hierarchical FM-index for graph-based alignment(Grytten et al. 2019). PanGenie (Elber et al. 2022) and KAGE (Grytten et al. 2022) utilize k-mer comparisons to haplotypes for rapid genotyping with reduced bias (Mun et al., 2023, Singh et al. 2022). rowbowt introduces a marker array approach that is faster and less memory-intensive than previous methods like BayesTyper (Crysnanto et al. 2022). GraphTyper realigns reads to variation subgraphs for population-scale genotyping(Eggertsson et al. 2017). Annotation of the pangenome holds the key for the functional importance of the pangenomes and to eluciddate the functional convergence and divergence of the pangenome. Tools such as ggCaller integrates gene prediction and pangenome clustering using DBGs, offering a ~50x speed increase over traditional workflows like Prokka/Roary (Horsfield et al. 2023). GrAnnoT and PPanG are specialized for annotation transfer and nucleotide-level visualization of genomic variations within graph-based browsers (Marthe et al. 2025, Liu et al., 2024).

Comparative studies reveal significant trade-offs between speed, memory, and graph complexity. mdbg is reported as the fastest construction tool with the lowest memory footprint, followed by Bifrost (Andreace et al. 2023). In contrast, PGGB and Minigraph-Cactus are more computationally demanding; PGGB has been noted to fail on very large datasets due to high resource requirements (Andreace et al. 2023). vg Giraffe is over an order of magnitude faster than GraphAligner for long reads (Chang et al. 2025). rowbowt consistently outperforms PanGenie and BayesTyper in terms of speed and memory efficiency (Crysnanto et al. 2022). ggCaller significantly reduces runtime compared to the Roary or Panaroo pipelines, scaling with graph complexity rather than the number of samples (Horsfield et al. 2023).

The next major limitation is functional interpretation. Identifying a genomic sequence is not equivalent to understanding its biological function. The growing number of novel genes, structural variants and regulatory regions identified through pangenomics creates a new problem: functional annotation is becoming a bottleneck.This suggests that the future of pangenomics should involve integration with multiple layers of biological information.Graph-based methods generally outperform linear reference methods in terms of accuracy and variant detection sensivitity, particularly for complex variant types. The VG toolkit demonstrates superior SV genotyping accuracy compared to tools like SVTyper and Paragraph (Hickey et al. 2019). Minigraph-Cactus and PGGB capture approximately 40% more small variations than Minigraph’s approximate mapping approach (Crysnanto et al. 2022, Leonard et al. 2023). vg shows higher sensitivity for short-read mapping on reads containing variants compared to HISAT2 and linear mappers (Grytten et al. 2019, Outten and Warren 2021).

Panscape : Supporting integrated pangenomics

Panscape is implemented as a single Rust command-line binary available from Codeberg (https://codeberg.org/gsablok/panscape) providing 24 subcommands (Figure 3, Supplementary Table 1) covering an end-to-end pangenomics support for miscellaneous pangenomics task for which user write custom scripts: read/FASTQ preprocessing (format conversion, length filtering, motif and regex scanning, adapter/tag clipping), alignment and annotation (minimap2-based read alignment, GFF-based CDS/mRNA extraction, miniprot-derived CDS extraction), PAF- and GFA-based pangenome comparison (alignment summarization, cross-alignment matching, graph summarization, region extraction), VCF-based variant comparison, and an assembly/database pipeline (hifiasm → compleasm → miniprot, with SQLite-backed read storage), paf alignment to PAF based graph for visualization of the pangenome graphs. Panscape is assisted with bash based entire pipeline for pangenome construction and annotation longpan (https://codeberg.org/gsablok/longpan), and graphpan (https://codeberg.org/gsablok/graphpan). Panscape is used as an ancillary tool for the pangenomics to interpret and summarize the pangenomics output. Each subcommand accepts an explicit thread-count parameter that controls an internal rayon-based parallel pool. Functional correctness was validated against a bundled sample-file set comprising GFA graphs, PAF alignments, paired FASTA files, and a real PacBio HiFi FASTQ read. For commands with pre-computed reference outputs (graph summarization, cross-PAF matching, and PAF-guided FASTA region extraction), output was confirmed byte-identical to the reference via checksum comparison.

To evaluate the effect of a set of implementation optimizations (buffered I/O, a corrected FASTA-parsing routine, rayon-based parallelization, and several algorithmic complexity reductions), we benchmarked an “original” and an “optimized” build of the same codebase, compiled with the same toolchain (rustc 1.75) and default Cargo release profile. Synthetic datasets of 20,000 bp reads were generated at three sizes (15, 50, and 100 MB; https://codeberg.org/gsablok/panscape) in both FASTA and FASTQ format, the latter with randomly generated Phred-like quality strings. Each command/size/format combination was timed as the best of three wall-clock runs (date +%s.%N), with output redirected to /dev/null and --thread 1 throughout. Because the benchmarking environment had a single CPU core (nproc = 1), these measurements isolate the non-threading optimizations only; parallel speedups from multi-threaded execution on multi-core hardware were not evaluated.

Across 18 format/size/command combinations, the optimized build was faster than the original in 17 of 18 cases, with speedups ranging from 1.05× to 2.32× (Supplementary Table 2). Gains were largest for scanner and stat on smaller FASTA inputs (2.32× and 1.67× at 15 MB, respectively) and remained above 1.2× for most commands at 100 MB. A single regression was observed: filterreads on 100 MB FASTQ input was approximately 6% slower in the optimized build (0.121 s vs. 0.128 s), and showed only marginal improvement (1.05×) at 50 MB. This result was consistent across five repeated runs, indicating a reproducible effect rather than measurement noise. Thus, panscape provides an optimized and scalabale workflow integration with several pangenomics outputs from several pangenome graph aligners and variant calling procedures (Table 1), which has matured significantly with graph based approaches establishing themselves as the new paradigm for the genomic analysis. While computational overhead and visualization remain significant hurdles, the shift towards reference-agnostic representations is providing unprecedented insights into the genetic diversity of species under study in pangenomics. Continued efforts in standardization and algorithmic optimization will be critical for the widespread adoption of pangenomics in routine genomic workflows.

Table 1. Comparative analysis of pangenome software and panscape.

Feature	panscape	PGGB	Minigraph-Cactus	Minigraph	Panaroo
Primary purpose	Integrated pangenomic workflow	Reference-free pangenome graph construction	Pangenome graph construction	Fast pangenome graph construction	Gene-centric bacterial pangenomics
Language / implementation	Rust	Pipeline integrating multiple tools	Cactus + Minigraph ecosystem	C/C++-based tool	Python
Raw FASTQ/FASTA processing	Yes	Limited; mainly sequence preparation	Not primary focus	Not primary focus	Primarily assembled genomes/annotations
Motif searching	Yes	No	No	No	No
Read filtering/clipping	Yes	No	No	No	No
PAF analysis	Yes	Uses PAF internally	Yes, through components	Yes	Not its main focus
GFA graph analysis	Yes	Core function	Core function	Core function	GML-based gene graph
Genome annotation	Yes	Limited/downstream	Limited/downstream	Limited	Core function
VCF analysis	Yes	Yes, through vg-related workflow	Yes	Yes/downstream	Not primary
Ancestral-state analysis	Yes	Not a primary feature	Not a primary feature	Not a primary feature	Limited
Read database	Yes	No	No	No	No
Integrated single executable	Yes	No; multiple components	No; multiple components	More focused	Primarily pipeline/software
Main strength	Breadth and integration	High-quality reference-free graphs	Complex, scalable genome graphs	Computational efficiency	Gene presence/absence and bacterial genomes

Table 1. Comparative analysis of pangenome software and panscape.

Feature	panscape	PGGB	Minigraph-Cactus	Minigraph	Panaroo
Primary purpose	Integrated pangenomic workflow	Reference-free pangenome graph construction	Pangenome graph construction	Fast pangenome graph construction	Gene-centric bacterial pangenomics
Language / implementation	Rust	Pipeline integrating multiple tools	Cactus + Minigraph ecosystem	C/C++-based tool	Python
Raw FASTQ/FASTA processing	Yes	Limited; mainly sequence preparation	Not primary focus	Not primary focus	Primarily assembled genomes/annotations
Motif searching	Yes	No	No	No	No
Read filtering/clipping	Yes	No	No	No	No
PAF analysis	Yes	Uses PAF internally	Yes, through components	Yes	Not its main focus
GFA graph analysis	Yes	Core function	Core function	Core function	GML-based gene graph
Genome annotation	Yes	Limited/downstream	Limited/downstream	Limited	Core function
VCF analysis	Yes	Yes, through vg-related workflow	Yes	Yes/downstream	Not primary
Ancestral-state analysis	Yes	Not a primary feature	Not a primary feature	Not a primary feature	Limited
Read database	Yes	No	No	No	No
Integrated single executable	Yes	No; multiple components	No; multiple components	More focused	Primarily pipeline/software
Main strength	Breadth and integration	High-quality reference-free graphs	Complex, scalable genome graphs	Computational efficiency	Gene presence/absence and bacterial genomes

Existing pangenomic software frequently focuses on individual stages of the workflow, particularly pangenome graph construction, genome alignment, or gene-level presence/absence analysis. In contrast, panscape provides an integrated command-line framework encompassing read preprocessing, motif analysis, alignment processing, genome annotation, PAF and VCF comparison, graph summarization, ancestral-state analysis, and pangenome database construction. This broader functional scope distinguishes panscape from specialized graph-construction frameworks such as PGGB and Minigraph-Cactus and gene-centric approaches such as Panaroo. Overall, panscape represents an integrated and performance-oriented computational framework for pangenomic analysis, combining read preprocessing, sequence analysis, alignment comparison, genome annotation, graph analysis, and statistical summarization within a single Rust executable. Existing pangenomic computational frameworks, including PGGB, Minigraph, Minigraph-Cactus, and Panaroo (Tonkin-Hill et al. 2020), are generally designed to address specific aspects of pangenome analysis. PGGB (Garrison et al. 2024), Minigraph, and Minigraph-Cactus (Hickey et al. 2024) primarily emphasize genome alignment and pangenome graph construction, whereas Panaroo is principally oriented toward gene-level analysis, particularly gene clustering and presence–absence characterization in bacterial genomes. In contrast, panscape integrates a broader range of analytical functions within a single computational framework, encompassing FASTA/FASTQ preprocessing, read filtering and sequence clipping, motif detection, alignment analysis, genome annotation, PAF and VCF comparison, GFA graph summarization, ancestral-state analysis, and pangenome database construction. The principal distinction of panscape therefore lies in its functional breadth and workflow integration, rather than in specialization in a particular component of pangenomic analysis.