Accession naming should adhere to MIAPPE (Minimum Information About Plant Phenotyping Experiments) standards. The Biological Material ID should incorporate institutional identifiers, followed by the accession number from germplasm catalogue or common name of the plant source/variety (e.g., IPK-Gatersleben:HOR_13170 for barley accession “Barke”) to ensure traceability (MIAPPE v1.1, Papoutsoglou et al., 2020). When complementary data regarding a specific accession is also available at external sources (e.g. Biosamples), a link to a Biological material external ID should be provided in the metadata.

Genome assembly identifiers should contain at least 4 fields—species, variety/line, project group, assembly version — separated by period (‘.’), with an optional fifth field for additional information ( Cannon et al., 2025). For example, drOrySati.Nipponbare.RicePan.1.0, which refers to the assembly of Oryza sativa, Nipponbare cultivar, RicePan project, version 1.0 (ToLID identifier, https://id.tol.sanger.ac.uk/, Darwin Tree of Life Consortium, 2023).

Gene identifiers must balance stability with biological relevance, as outlined by Cannon et al. (2025), keeping track of the annotation version, chromosome and gene ID. Their framework proposes human- and machine-readable identifiers, including the assembly names (e.g. drOrySati.Nipponbare.RicePan.1.0) with the addition of gene models like drOrySati.Nipponbare.RicePan.1.0.1.01.g000100 (assembly version 1.0, annotation version 1, chromosome 01, gene 100). To enhance this for pan-genomics, the “group” field can denote pan-genome projects (e.g., RicePan), linking multiple assemblies, while optional fields like “Hap1” or metadata tags distinguish haplotypes or accession types (e.g., wild vs. cultivated). Pangenes, representing orthologous gene clusters, can be assigned identifiers like drOrySati.RicePan.pan00001, with metadata linking to specific gene models across assemblies. Cannon et al. (2025) advocate preserving legacy identifiers via cross-references to ensure stability, avoiding disruptive renaming as new accessions are added.

A core metadata schema is critical for interoperability. Required fields to properly annotate pan-genome studies include species details such as name (TaxonID), pedigree, geographic origin, ploidy and chromosome number, as well as sequencing technology used (e.g. PacBio HiFi, Oxford Nanopore, Hi-C, Illumina), assembly pipelines (e.g., Flye ( Kolmogorov et al., 2020), hifiasm ( Cheng et al., 2024), Canu ( Koren et al., 2017), …), and assembly QC metrics (e.g., BUSCO scores ( Manni et al., 2021)). Existing ontologies such as the Sequence Ontology (SO) should be extended to include pan-genome-specific terms that describe the layouts and structures of pan-genomes ( Eilbeck et al., 2005). These can be categorized as core, shell and cloud genome genes, but these terms may depend on the number of genomes and genotypes selected ( Jayakodi et al., 2024). Any downstream comparative analysis requires open and transparent reporting on the thresholds used, so that these must be included in the metadata. Collaboration with the AgBioData Nomenclature Working Group and the Genomics Standards Consortium ( https://www.gensc.org/) ensures alignment with broader genomic standards ( Cannon et al., 2025).

As pan-genome graphs grow to encompass not just core and variable genes but a full spectrum of genomic elements, we need a unified identification system. Current annotation often focuses on genes, leaving features like transposable elements, SSRs, non-coding RNAs, and regulatory motifs with inconsistent or tool-specific labels. We propose the development of a generalized feature identifier (GFI). This system would provide a stable, queryable, and standardized format for any annotated feature, independent of its type or the discovery tool used. A GFI would be important for pan-genome-scale association studies and for functionally characterizing the entire “dark matter” of the genome, ensuring that a SNP in a long terminal repeat or a copy number variation in a novel ncRNA can be cataloged and compared with the same rigor as in a protein-coding gene.

Quality control in genome assembly workflows begins with sequencing QC, where tools like FASTQC assess raw read integrity, including base quality, GC content, and adapter contamination ( Figure 1A). K-mer plots, generated via Jellyfish ( Marçais et al., 2011) paired with GenomeScope 2 ( Ranallo-Benavidez et al., 2020), provide insights into genome complexity, such as ploidy, heterozygosity, and repetitive element profiles ( Figure 1A). For individual assembly QC, QUAST ( Mikheenko et al., 2023) is recommended for evaluating contiguity metrics (e.g., N50, L50) and is particularly effective for comparing multiple assemblies of diploid genomes, while CRAQ ( Li et al., 2023) excels in assessing consensus accuracy and structural errors in polyploid genomes due to its sensitivity to haplotype-specific misassemblies ( Figure 1A). When results from QUAST and CRAQ conflict (e.g., differing contig counts due to haplotype collapsing), users should prioritize CRAQ for polyploid assemblies and cross-validate with raw read alignments (e.g., using Minimap2) to resolve discrepancies. Merqury ( Rhie et al., 2020) further validates haplotype resolution in polyploid or heterozygous genomes (e.g., wheat, potato) by comparing k-mer spectra between raw reads and assemblies, offering a robust check for completeness and phasing errors ( Figure 1A). For repeat quality control, the LTR Assembly Index (LAI; Ou et al., 2018) assesses the completeness of long terminal repeat retrotransposons, while tidk ( Brown et al., 2025) detects telomeric motifs to evaluate chromosomal end-to-end integrity ( Figure 1A). When results from these tools conflict, LAI generally provides a more reliable indicator of assembly quality. Even in high-quality plant genomes assembled from long reads, some chromosome ends may still lack detectable telomeric repeats.

Annotation pipelines must be documented alongside assembly strategies. These may include gene model integration pipelines like MAKER2 ( Holt & Yandell, 2011), PASA ( Haas et al., 2003) or BRAKER3 ( Gabriel et al., 2024), while Helixer ( Stiehler et al., 2020) is recommended for ab initio prediction in non-model organisms due to its deep learning-based approach ( Figure 1B). Liftoff ( Shumate & Salzberg, 2021) is ideal for annotation transfer between closely related species and should be part of a standard annotation pipeline ( Figure 1B). Use versioned workflows (e.g., Snakemake ( Köster & Rahmann, 2012), or Nextflow ( Di Tommaso et al., 2017)) to ensure reproducibility, provenance tracking, and portability. Transcriptomic data (RNA-Seq) from multiple tissues (e.g., roots, shoots) and stress conditions (e.g., drought, disease) with sufficient read coverage validates gene models, especially for accessory genes lacking orthologs ( Qin et al., 2021). Long-read RNA sequencing technologies are recommended to recover full-length transcripts and accurately characterize alternative isoforms. For structural annotation QC, BUSCO ( Manni et al., 2021) assesses gene space completeness using lineage-specific datasets that can be adjusted for polyploid genomes ( Figure 1B). Mercator4 ( Bolger et al., 2021) assigns functional categories based on the MapMan bin system and is useful for identifying missing functions in a single genome ( Figure 1B). PSAURON ( Sommer et al., 2025) validates structural annotations, and OMArk ( Nevers et al., 2025) detects contamination via evolutionary consistency checks ( Figure 1B). In cases where evaluation tools disagree (e.g., BUSCO reports missing genes but PSAURON suggests completeness), integrating RNA-Seq support and orthology evidence provides a more reliable basis for resolving such discrepancies.

Pan-genome completeness requires saturation analysis, where gene accumulation curves assess whether additional accessions contribute novel genes ( Tettelin et al., 2005). For species with varying ploidy levels (e.g., diploid vs. polyploid barley), a minimum of 10–20 accessions is typically required for diploid species to approach saturation, while polyploid species may need 30–50 accessions due to increased gene content complexity ( Jayakodi et al., 2024). Users should plot accumulation curves using tools like Panaroo and evaluate saturation by fitting models (e.g., Heap’s Law) to confirm diminishing returns in gene discovery ( Figure 1C). For species like barley, benchmark datasets of 100+ conserved genes enable orthology tool validation ( Jayakodi et al., 2024). OrthoFinder ( Emms & Kelly, 2019) and CAFE5 ( Mendes et al., 2020) facilitate gene family expansion and contraction analyses, providing insights into evolutionary dynamics ( Figure 1C). Structural variant detection, using Sniffles2 ( Smolka et al., 2024) for long-read data or SVIM ( Heller & Vingron, 2019) for short-read data, quantifies indels and inversions ( Qin et al., 2021) ( Figure 1C). When tools like Sniffles2 and SVIM yield conflicting variant calls, users should integrate multi-platform data (e.g., combining long- and short-read alignments) and prioritize calls supported by higher read depth or mapping quality. Presence-absence variation (PAV) detection via Panaroo or PAV-specific pipelines is critical for identifying variable gene content tied to phenotypic diversity ( Tonkin-Hill et al., 2020).

•

Raw data: Assemblies must be submitted in FASTA format with headers containing unique sequence identifiers (e.g., >chr01, >chr02). Annotations must be provided in GFF3 or GTF format (compliant with Sequence Ontology), with the sequence IDs in the first column exactly matching the sequence identifiers used in the FASTA headers.

Centralized repositories would archive versioned datasets (e.g., Barley v2, Rice v1.5) with DOI-based identifiers (DataCite). Public deposition in INSDC (raw reads and assembly, https://www.insdc.org/) and Ensembl (annotations, see documentation of Ensembl, 2025, https://beta.ensembl.org/) ensures global accessibility (ENA Documentation, 2025).

Mandatory metadata fields include sequencing technology and coverage (e.g., PacBio HiFi, Oxford Nanopore), assembly method (e.g., Flye, Hifiasm), accession provenance (BioSample IDs), and software versioning of all software and pipelines used. Missing metadata, as observed in early barley submissions, must be addressed via enforced submission guidelines ( Jayakodi et al., 2024).

For pangenome datasets, additional metadata fields are critical to ensure traceability and interoperability across studies. These should include the species name and NCBI Taxonomy ID, pangenome version and build date, and a complete list of constituent genomes with corresponding assembly accessions, strain names, and versions. Furthermore, metadata should describe the methods and parameters used to construct the pangenome.

Capturing this information in structured formats such as JSON-LD or RO-Crate ( Peroni et al., 2022) would align pangenome submissions with broader FAIR data principles and facilitate integration with knowledge graphs and comparative genomics resources.

Plant pan-genomes capture a species’ full genomic diversity, constructed using either linear-based or graph-based methods, each with distinct strengths and limitations. To provide a clearer comparison, linear-based approaches are divided into two distinct categories: sequence-based and gene-based analyses.

Sequence-based linear analysis involves aligning multiple genomes to a single reference or consensus sequence to identify sequence-level variations, such as single-nucleotide polymorphisms (SNPs) and insertions/deletions (indels). This process typically employs variant callers like GATK ( McKenna et al., 2010) or freebayes ( Garrison & Marth, 2012) to detect SNPs and indels from whole-genome alignments. These methods are computationally efficient and compatible with visualization tools like JBrowse2 ( Diesh et al., 2023) or IGV ( Robinson et al., 2023) for synteny and variant visualization ( Figure 1D). Web-portals such as PanBARLEX ( PanBARLEX - Barley Pangenome Explorer) enable pan-genome research by providing searchable and pre-rendered visualizations ( Figure 1E). However, reference bias in sequence-based linear approaches can limit their ability to capture complex structural variations, particularly in repetitive or polyploid plant genomes.

Gene-based linear analysis focuses on inferring orthology and identifying gene-level presence/absence variations (PAVs) using tools like OrthoFinder ( Emms & Kelly, 2019) or Ensembl Compara ( Dyer et al., 2025). These tools analyze annotated gene sets to determine the pan-gene repertoire, identifying core and accessory genes across a species. While effective for gene-level PAV detection, these methods do not directly address sequence-level variations like SNPs or indels, requiring separate workflows for comprehensive analysis. Orthology inference tools must be benchmarked using inflation value sweeps to minimize false positives ( Emms & Kelly, 2019). Visualization of gene-level PAVs can be achieved through UpSet plots or as presence/absence relationships in KnetMiner knowledge graphs ( Hassani-Pak et al., 2021) ( Figure 1F).

In contrast, graph-based approaches model genomes as interconnected nodes (shared regions) and edges (SNPs, indels, and structural variants) using tools like VG Toolkit ( Hickey et al., 2020), PGGB ( Garrison et al., 2024), PanTools ( Jonkheer et al., 2022), or wfmash ( Guarracino et al., 2021) ( Figure 1D). These methods integrate both sequence-level and structural variations in a single framework, offering an unbiased, comprehensive view of genomic diversity. They are particularly suited for complex genomes, such as tomato ( Zhou et al., 2022). Visualization tools like Bandage ( Wick et al., 2015) or Cytoscape ( Shannon et al., 2003) are used to represent structural complexity, though these approaches are computationally intensive and require specialized expertise ( Figure 1D).

In summary, sequence-based linear methods excel in rapid SNP and indel detection but are limited by reference bias, while gene-based linear methods are ideal for pan-gene analysis but require separate homology-based workflows. Graph-based approaches unify both gene-level and structural variation analyses, offering greater flexibility for complex genomes despite higher computational demands. As computational resources and tools advance, graph-based methods are becoming more accessible, enhancing plant pan-genome studies as demonstrated in rice ( Qin et al., 2021).

The integration of pangenomic information into crop improvement remains challenging, despite its potential to illuminate the genetic basis of agronomic traits. Pangenomes reveal extensive structural polymorphisms and gene content diversity across accessions, yet these findings often remain siloed from other key data sources such as GWAS and QTL mappings, gene expression profiles, gene regulation, functional annotations, and published literature. Without coherent integration, researchers face difficulties in linking genomic variation to phenotype and in distinguishing biologically meaningful signals from background noise. Bridging these data types requires frameworks capable of harmonizing heterogeneous evidence, tracking provenance, and enabling transparent reasoning across molecular, phenotypic, and bibliographic domains.

Platforms such as KnetMiner ( Hassani-Pak et al., 2021, https://knetminer.com) address these challenges by synthesizing pangenomic, association, omics, and literature-derived evidence within a unified knowledge graph. This integrative approach allows relationships among genes, traits, and pathways to be explored in context, supporting AI-assisted hypothesis generation and candidate gene prioritization. By providing explainable connections between diverse evidence sources, KnetMiner exemplifies how knowledge graph technologies can transform FAIR yet fragmented genomic data into a coherent foundation for evidence-based crop breeding.