Cracking Open the Green Genome: A Field Guide to Plant Genome Assembly
Cracking Open the
Green Genome
A field guide to the tools, algorithms, and biological challenges shaping modern plant genome assembly
Plants are genomic marvels — and genomic nightmares. A single wheat genome stretches across 16 billion base pairs, roughly five times the size of the human genome, and harbors a labyrinth of repetitive sequences, polyploid layers, and structural rearrangements that have confounded sequencers for decades. Yet in the span of just a few years, a convergence of long-read sequencing chemistry, graph-based assembly algorithms, and chromosome-scale scaffolding has transformed plant genomics from a brute-force endeavor into something closer to precision engineering. Understanding how these technologies work — and where they still fall short — is essential for any bioinformatics practitioner navigating the green frontier.
Why Plant Genomes Are Hard
The challenge of plant genome assembly is not simply one of size. It is a tangle of biological properties that conspire against short-read sequencing. Three forces dominate: polyploidy, transposable element proliferation, and heterozygosity.
Polyploidy: the copies problem
Many crop plants are polyploids — they carry two or more complete sets of ancestral subgenomes. Bread wheat is allohexaploid, meaning it contains three distinct but related subgenomes (A, B, and D) inherited from three ancestral grass species. When a short read lands on a gene in the A subgenome, it may be indistinguishable from its homeologous counterpart in B or D. The assembler collapses what should be three separate loci into one, producing a chimeric consensus that misrepresents all three. This problem of homeologous collapse plagued the first wheat drafts and was only resolved once long-read technologies allowed reads to span the diagnostic variants that distinguish subgenomes.[1]
Transposable elements: the repetitive flood
In maize, roughly 85% of the genome is composed of transposable elements (TEs) — mobile DNA sequences that have replicated and inserted themselves throughout the genome over millions of years. Long terminal repeat retrotransposons (LTR-RTs) alone account for more than 75% of the maize genome. When an assembler encounters reads that could originate from any of thousands of identical TE copies, it cannot determine where each read belongs. The result is fragmented assemblies riddled with gaps precisely where TE clusters occur — and TE clusters coincide with centromeres, pericentromeric heterochromatin, and the flanking regions of many agronomically important genes.[2]
Heterozygosity: diploid ambiguity
Outcrossing plant species like potato, cassava, and many tree species are highly heterozygous — the two copies of each chromosome differ substantially. A short-read assembler collapses heterozygous loci into a single consensus, losing allele-specific information that may be critical for trait mapping. Phased, haplotype-resolved assemblies are therefore not a luxury in plant genomics; they are a scientific necessity for species where allelic variation drives phenotypic diversity.
The Sequencing Revolution: HiFi and ONT
Two long-read technologies have fundamentally changed what is possible in plant genome assembly: PacBio HiFi (High-Fidelity) sequencing and Oxford Nanopore Technologies (ONT) sequencing.
HiFi reads are generated by passing a single DNA molecule around a circular template multiple times, producing a consensus read that averages the errors from each pass. The result is reads of 15–25 kb with accuracy exceeding 99.5% — long enough to span most TE insertions, accurate enough to be used directly by assembly algorithms without error correction. For plant genomes, HiFi has proven transformative: the first chromosome-scale assembly of bread wheat leveraging HiFi reduced fragmentation by an order of magnitude compared to prior short-read and error-prone long-read assemblies.[3]
ONT ultra-long reads push even further — recent library preparations yield reads exceeding 100 kb, and runs targeting ultra-long DNA have produced reads of 1–4 Mb. These reads can span entire TE arrays, resolve centromeric satellite repeats, and bridge gaps that even HiFi cannot cross. The trade-off is a base error rate of 3–5% in standard mode, though ONT's duplex chemistry and R10 pore chemistry have brought accuracy close to HiFi levels. In practice, hybrid assemblies combining HiFi for accuracy and ONT ultra-long for repeat resolution have produced some of the most complete plant genome assemblies to date.[4]
Assembly Algorithms: From Overlap-Layout-Consensus to Hifiasm
The algorithmic history of genome assembly roughly maps onto the read technologies of each era. Early short-read assemblers used de Bruijn graphs — a representation that decomposes reads into fixed-length k-mer substrings and finds Eulerian paths through the graph. This approach scales efficiently to millions of short reads but collapses under repetitive sequences longer than k.
Long-read assemblers have largely migrated to overlap-layout-consensus (OLC) frameworks and string graphs. In this paradigm, all reads are compared pairwise to find overlaps, an assembly graph is constructed from those overlaps, and consensus sequences are computed for each path through the graph. The key insight is that long reads can bridge repetitive regions whose internal sequence is ambiguous, because the unique flanking sequences on either side anchor the read to a unique position in the assembly graph.
Hifiasm, released in 2021 and continuously updated, has become the de facto standard for HiFi-based plant assembly. Its key innovation is phasing-aware graph construction: Hifiasm uses heterozygous SNPs within reads to partition overlaps into haplotype-specific paths, producing two fully phased assembly graphs simultaneously — one per haplotype — without requiring a reference genome or parental data. For allotetraploid and allohexaploid plants, extensions of this approach using Hi-C or trio binning data allow haplotype graphs to be separated into distinct subgenome assemblies.[5]
| Tool | Input | Graph Type | Key Feature | Best For |
|---|---|---|---|---|
| Hifiasm | HiFi (±Hi-C) | String / overlap | Native haplotype phasing | Diploid / polyploid |
| HiCanu | HiFi / CLR | OLC | Robust repeat resolution | Highly repetitive |
| Verkko | HiFi + ONT | de Bruijn + OLC | Telomere-to-telomere | T2T assemblies |
| Flye | ONT / HiFi | Repeat graph | Highly scalable; handles ultra-long ONT | Large genomes |
| wtdbg2 | ONT / CLR | Fuzzy Bruijn graph | Speed; memory-efficient | Large genomes fast |
| IPA | HiFi | OLC | PacBio-native pipeline | Mid-size plants |
Scaffolding to Chromosome Scale
Even the best contig-level assembly remains a collection of fragments unless it can be ordered and oriented into chromosome-scale pseudomolecules. Three complementary scaffolding technologies dominate modern workflows.
Hi-C chromatin proximity ligation
Hi-C captures the three-dimensional organization of chromatin in the nucleus. Genomic regions that are physically close in space — and are therefore on the same chromosome, or on chromosome arms that contact each other during interphase — are preferentially ligated. The resulting paired-end reads, when mapped to contigs, reveal which contigs are likely neighbors along a chromosome. Tools like 3D-DNA, SALSA2, and YaHS use Hi-C contact maps to order and orient contigs into chromosome-scale scaffolds with remarkably high accuracy. For plants, Hi-C is now essentially standard — nearly every published chromosome-scale plant genome assembly since 2019 has used it.[6]
Optical mapping
Optical mapping technologies (Bionano Genomics) label long DNA molecules at specific restriction sites or sequence motifs and image them in nanochannels to produce a physical map of label positions across molecules averaging 150–300 kb. These maps provide an orthogonal source of distance information that can validate or correct Hi-C-based scaffolding, and excel at detecting large structural variants (>10 kb) that confound sequence-based approaches.
Genetic maps and synteny
Traditional genetic linkage maps derived from recombinant inbred lines or F2 populations remain valuable for validating chromosome assignments. Comparative synteny with closely related reference genomes provides a "sanity check" — if a newly assembled contig is syntenic with a known chromosome in a relative, its assigned chromosome in the new assembly should match. Tools like MCScan and JCVI's synteny suite enable rapid cross-species chromosome validation.
Genome Annotation: Finding the Genes
A chromosome-scale assembly without annotation is an atlas with no place names. Structural annotation — identifying protein-coding genes, non-coding RNAs, and repetitive elements — is a major computational undertaking in plants because gene space is interspersed with enormous TE arrays. The standard workflow involves three evidence layers: ab initio gene prediction using tools like AUGUSTUS or SNAP trained on plant-specific gene models; transcript evidence from RNA-seq or ISO-seq (full-length transcript sequencing); and protein homology from closely related species. MAKER and BRAKER2 are the most widely used integrative annotation pipelines for plants, combining all three evidence streams into a consensus gene set.[7]
TE annotation is equally critical. Unmasked TEs contaminate gene predictions and distort evolutionary analyses. RepeatModeler builds a species-specific TE library de novo, which RepeatMasker then uses to soft-mask the genome (replacing TE sequence with lowercase letters) before gene prediction. For polyploids, subgenome-specific TE content is an active research area — different subgenomes often harbor distinct TE families that can serve as phylogenetic markers for subgenome assignment.
| Species | Ploidy | Genome Size | Technology | Assembly Quality | Significance |
|---|---|---|---|---|---|
| Bread Wheat (IWGSC RefSeq v2) | Hexaploid (6×) | 14.5 Gbp | HiFi + Hi-C | 99.7% BUSCO | First fully phased 3-subgenome assembly |
| Maize (B73 v6) | Diploid | 2.4 Gbp | HiFi + ONT + Hi-C | T2T gapless | Centromere & knob resolution |
| Strawberry (F. × ananassa) | Octoploid (8×) | ~800 Mbp | HiFi + Hi-C | 8 subgenomes phased | Most complex polyploid crop assembled |
| Potato (DM v6 + RH) | Autotetraploid (4×) | ~1.7 Gbp | HiFi + Hi-C + BioNano | 4 haplotypes | First 4-haplotype resolved tuber crop |
| Arabidopsis thaliana (Col-CC) | Diploid | 135 Mbp | HiFi + ONT | T2T complete | First plant T2T, including all 5 centromeres |
| Sugarcane (R570 mono.) | Complex polyploid | ~10 Gbp | HiFi + Hi-C | In progress | Most agronomically complex genome |
Quality Assessment: BUSCO and Beyond
Benchmarking Universal Single-Copy Orthologs (BUSCO) has become the lingua franca of assembly quality. BUSCO queries a genome assembly against a curated set of genes expected to be present exactly once in any complete genome of a given lineage. A wheat assembly might be assessed against the embryophyta_odb10 dataset (~1,614 orthologs); a high-quality assembly typically reports >98% complete BUSCOs. But BUSCO completeness measures gene space, not repeat space — an assembly can have near-perfect BUSCO scores while leaving centromeres and pericentromeric heterochromatin entirely unresolved.[8]
Complementary metrics include: N50/NG50 (the contig length at which 50% of the assembly or expected genome size is covered — longer is better); LAI (LTR Assembly Index), which evaluates the intactness of LTR retrotransposons as a proxy for repeat resolution quality; and QV (quality value), estimated by mapping unassembled reads back to the assembly and counting disagreement positions — a QV of 50 corresponds to one error per 100,000 bases, the current threshold for "reference quality."
The Telomere-to-Telomere Horizon
In 2022, the Telomere-to-Telomere (T2T) consortium published a complete, gapless human genome assembly — every chromosome from telomere to telomere, including all five previously unresolved centromeres. For plants, the T2T milestone was reached for Arabidopsis thaliana shortly thereafter, resolving all five centromeres and revealing that centromeric satellite arrays are far more structurally diverse and gene-poor than anticipated.[9]
Achieving T2T status requires ONT ultra-long reads to span the megabase-scale satellite arrays of centromeres, HiFi reads to resolve the base-level sequence within those arrays, and specialized assemblers (Verkko, in particular) that can thread through graphs containing thousands of nearly-identical repeat units. For crop plants with complex polyploid genomes, true T2T assembly remains a frontier — each haplotype of each subgenome requires independent centromere resolution, multiplying the complexity by ploidy level. The T2T wheat genome — all three subgenomes, all 21 chromosome pairs, all centromeres resolved — remains one of the most ambitious targets in plant bioinformatics.[10]
What T2T Unlocks for Agriculture and Beyond
Complete, phased plant genome assemblies are not merely academic achievements. They enable a cascade of downstream applications with direct agronomic and economic relevance. Fully resolved centromeres clarify meiotic crossover suppression, explaining why certain genomic intervals are recalcitrant to breeding — knowledge that breeders can use to design crossing schemes that break undesirable linkage blocks. Phased haplotype assemblies allow genome-wide association studies (GWAS) to operate at haplotype resolution, identifying causal variants that are invisible to pseudomolecule-based analyses. Pan-genome projects — assembling dozens to hundreds of accessions of the same species — are revealing the "dispensable" genome: structural variants, presence-absence polymorphisms, and inversions that explain phenotypic diversity within a species but are absent from any single reference.
For the bioinformatics consultant, these developments translate into a rapidly growing demand for genome assembly pipelines, annotation workflows, comparative genomics analyses, and pan-genome infrastructure. The tools exist; the bottleneck is expertise in threading them together for species whose genomic properties defy off-the-shelf solutions.
Conclusion
Plant genome assembly has undergone a revolution so rapid that textbooks written five years ago are already historical documents. The combination of HiFi sequencing, ONT ultra-long reads, Hi-C scaffolding, and phasing-aware assembly algorithms has made chromosome-scale, haplotype-resolved assemblies achievable for almost any plant species — including the largest and most complex polyploid crops. The telomere-to-telomere frontier is being actively mapped, and pan-genome projects are beginning to capture the true breadth of genomic diversity within species. What remains is the hard work of annotation, interpretation, and translation — turning assembled sequence into biological insight and agronomic utility. That is where bioinformatics expertise matters most.
References
- Appels R. et al. / IWGSC (2018). Shifting the limits in wheat research and breeding using a fully annotated reference genome. Science, 361(6403), eaar7191.
- Jiao Y. et al. (2017). Improved maize reference genome with single-molecule technologies. Nature, 546, 524–527.
- Cheng H. et al. (2021). Haplotype-resolved de novo assembly using phased assembly graphs with hifiasm. Nature Methods, 18, 170–175.
- Wick R.R. et al. (2023). Assembling the perfect bacterial genome using Oxford Nanopore and Illumina sequencing. PLOS Computational Biology, 19(3), e1010905.
- Dudchenko O. et al. (2017). De novo assembly of the Aedes aegypti genome using Hi-C yields chromosome-length scaffolds. Science, 356(6333), 92–95.
- Holt C. & Yandell M. (2011). MAKER2: an annotation pipeline and genome-database management tool for second-generation genome projects. BMC Bioinformatics, 12, 491.
- Simão F.A. et al. (2015). BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs. Bioinformatics, 31(19), 3210–3212.
- Michael T.P. & Jackson S. (2024). The first complete reference sequence for a flowering plant genome. The Plant Cell, 36(5), 1091–1108.
- Nurk S. et al. (2022). The complete sequence of a human genome. Science, 376(6588), 44–53.
- Dou J. et al. (2023). Haplotype-resolved genomes of octoploid cultivated strawberry reveal independent subgenome evolution and breeding history. Nature Plants, 9, 151–163.