Where we went, written in DNA

The story this dataset tells is, at first, a story about its own gaps. We assembled 19,029 ancient individuals from the Allen Ancient DNA Resource, each carrying a place, a date, a curated ancestry label and, often, a paternal and maternal haplogroup. Yet when we plot those samples through time, 99.6 percent of them are younger than 30,000 years, and more than two-thirds come from Europe. The very window we set out to prioritise — the Out-of-Africa dispersal between 70,000 and 30,000 years ago — contains just sixty-eight individuals, and the African homeland that launched every later migration is represented by fewer samples than a single medieval European cemetery. Any honest reconstruction must therefore separate what the genomes confirm from what we infer around their silences.

What the genomes confirm is remarkable enough. Unsupervised DBSCAN clustering on a combined geographic-and-temporal metric recovered 226 spatio-temporal populations whose haplogroup signatures align precisely with the known archaeological record. We can watch a single thread run from Çatalhöyük's farmers in Anatolia (Y-haplogroup G2a) out across Europe as the Linearbandkeramik, their agricultural package advancing in step with early-Holocene warming and largely overwriting the resident hunter-gatherers. We can watch a second thread gather on the Pontic-Caspian steppe — Khvalynsk, then Yamnaya, marked by R1b — and burst westward around 5,000 years ago on the back of wagons, horses and dairying, remaking the European gene pool a second time as Corded Ware and Bell Beaker. The same steppe engine turns east as Sintashta and Andronovo, carrying R1a-Z93 and chariot technology toward Central and South Asia. Each of these events surfaced not as an assumption but as a top-ranked edge in our migration-flow model, where an older population is linked to a younger, genetically similar, geographically reachable neighbour.

Beyond Eurasia, the founder signatures are equally clean. Every Indigenous American cluster, from Anzick-1 in Pleistocene Montana to the Classic Maya, carries the Beringian Q1b paternal lineage — the fingerprint of a small population that paused in a glacial refugium while sea levels were low, then spread the length of two continents once deglaciation opened the coastal and interior routes. In Remote Oceania, the Mariana Islands' Latte-period people and the settlers of Vanuatu carry East Asian O lineages, the maritime calling-card of the Austronesian expansion that outrigger canoes made possible. In each case the driver is legible: a climate door opening, a technology unlocking new range, or a demographic engine — farming's carrying capacity — pushing one people into another's land.

The deepest and most important migration, however, we can only see in silhouette. The Out-of-Africa dispersal left no in-window genomes in its African or Arabian source. We infer it instead from its descendants: Ust'-Ishim in Siberia and Zlatý kůň and Bacho Kiro in Europe, all near 45,000 years old, with the Oase individual preserving a Neanderthal ancestor only a handful of generations back. From these downstream points, and from the climate record of MIS 3 with its intermittently "green" Arabia, we reconstruct a dispersal threading humid corridors out of Africa during favourable pulses — a confident outcome resting on an inferred route.

The methodological honesty matters as much as the findings. We did not process the multi-gigabyte genotype matrix; our "genetic similarity" rests on curated labels and uniparental markers, and our climate layer is a date-keyed chronology rather than sample-measured cores. The clustering's resolution depends on a tuned space-time exchange rate. None of this overturns the signal, but all of it bounds our certainty. The clearest scientific message is therefore double-edged: ancient DNA has turned the last ten thousand years of Eurasian prehistory into something we can read almost like a ledger, while the older, tropical and African chapters — the ones that actually made us a global species — remain frustratingly, instructively blank.