Complete genome and pangenome
The first gap-free assembly of every human chromosome except Y (2022, Y added in 2023) and a growing pangenome that captures human diversity.
Open in the interactive tree →The 2003 reference still missed about 8 percent of the DNA. The Telomere-to-Telomere consortium closed the gaps with long-read sequencing and published a complete 3.055-billion-letter genome (T2T-CHM13, all chromosomes except Y) in Science on 31 March 2022, adding nearly 200 million letters; a complete Y chromosome followed in 2023. A single reference misses differences between people, so the Human Pangenome Reference Consortium builds a collection of high-quality genomes.
As of October 2026
The Human Pangenome Reference Consortium released its second data release in May 2025, with phased assemblies from 232 individuals, nearly five times more than release 1, and for the first time with samples sequenced by partners in Japan and Italy. Long-read sequencing now yields many complete chromosome assemblies, and Illumina advertises sequencing at about 200 US dollars per genome (a company claim). Even 232 people are still a small sample of humanity, so the pangenome does not yet represent all human diversity.
Open steps
- Assembling centromeres and repeats High AI leverageCentromeres, duplicated segments and other repeat-rich regions are still hard to assemble in every person; routine gap-free assemblies need better algorithms.
- Clinical use of graph references High AI leverageGraph-aware pipelines exist (DRAGEN multigenome, vg Giraffe), but independent multi-sample clinical validation of graph-based structural-variant calling is still missing.
- Sequencing missing populations Low AI leverageA few hundred people cannot represent humanity; cheap long-read sequencing and local labs are needed so under-represented ancestries are included.
Where AI could help
Medium AI leverage. AI already raises assembly and variant-calling accuracy, but who gets sequenced, consent and sequencing cost set the pace.
- Machine-learning error correction for long-read assemblies
- Pangenome-aware variant calling across hundreds of genomes
- Assembling hard repeat regions such as centromeres
- Predicting the effect of newly found structural variants
Shown so far
- In 2025 Genome Research published DeepPolisher, a transformer assembly polisher that roughly halves assembly errors; on 180 assemblies of the next HPRC release it gave a 54% error reduction. source