The entrance to the Altai cave in Siberia where a Denisovan fossil was found in 2008.
Image: Cheburgenator/Wikimedia

Buried deep within the DNA of Asian individuals is a genetic clue pointing to the existence of an unknown human ancestor. Remarkably, it wasn’t a human who reached this startling conjecture, but rather an artificially intelligent algorithm. Welcome to archaeology in the 21st century.

New research published last week in Nature Communications suggests a yet-to-be discovered hominin interbred with modern humans tens of thousands of years ago. This mystery species eventually went extinct, but an AI developed by researchers from the Institute of Evolutionary Biology (IBE) and several other European institutions found traces of its existence in the DNA of present-day people with Asian ancestry. A press release issued by the Centre for Genomic Regulation said it’s the first time deep learning has been used to explain human history, “paving the way for this technology to be applied in other questions in biology, genomics and evolution.”


The mystery hominin is likely a hybrid species of Neanderthals and Denisovans, according to the new research. Neanderthals, who lived in Europe, and Denisovans, who spread to Siberia, southeast Asia, and Oceania, were a closely related group of early humans, diverging from a common ancestor around 744,000 years ago. When anatomically modern humans (Homo sapiens) spilled into Eurasia from Africa, they commingled and interbred with both Neanderthals and Denisovans, which we know through genetic research. In addition to the generous amount of DNA left behind by the Neanderthals, scientists have extracted Denisovan DNA from a well-preserved finger bone found in a Siberian cave. Today, we find traces of these extinct species in the DNA of non-African humans, though only Asians retain genetic remnants of the Denisovans.

But as the new research suggests, modern humans, in addition to interbreeding with Neanderthals and Denisovans, interbred with a third, albeit unknown, species.

This idea isn’t new. In 2016, a genetics study co-authored by Mayukh Mondal from Tartu University and Jaume Bertranpetit from Pompeu Fabra University—both of whom were involved in the new study as well—offered evidence showing that indigenous Australasian populations from South and Southeast Asia “harbor a small proportion of ancestry from an unknown extinct hominin, and this ancestry is absent from Europeans and East Asians.” Their new study is an attempt to show where this “small proportion of ancestry” came from.


To that end, Mondal and Bertranpetit submitted evidence in the form of a demographic analysis, one churned by deep learning and sprinkled with some statistical analysis. Their algorithm devised and compared numerous complex demographic models to make predictions about the history of interbreeding events in Eurasia. To do so, the researchers fed the algorithm a healthy diet of whole genome sequences derived from both contemporary and ancient DNA, enabling the neural network to create a large set of possible demographic histories. A statistical analysis then calculated which of these histories were the most probable.

The team used deep learning to teach the algorithm to “predict human demographics” through “hundreds of thousands of simulations,” Òscar Lao, principal investigator at the Centro Nacional de Análisis Genómico and a co-author of the new study, said in a statement. “Whenever we run a simulation we are travelling along a possible path in the history of humankind. Of all simulations, deep learning allows us to observe what makes the ancestral puzzle fit together.”

The piecing together of this genetic jigsaw puzzle yielded the proposed “third archaic introgression,” meaning a third hybridization event among ancient humans (the other two being the genetic mixture between modern humans and Neanderthals and modern humans with Denisovans). Modern humans interbred with an unknown third species, and according to the models it’s either a Neanderthal-Denisovan hybrid, or an early offshoot of the Denisovan lineage.


That Neanderthals and Denisovans, who share a common ancestor, may have interbred is not an outrageous suggestion. Last year, scientists discovered the 90,000-year-old remains of one such hybrid individual—a girl with a Neanderthal mom and Denisovan dad. Due to a lack of evidence, scientists aren’t sure if this was a one-off thing, or if Neanderthals and Denisovans interbred on the regular. The new study bolsters the suggestion that the two extinct species bred often enough to produce a genetically distinguishable hybrid population.

Serena Tucci, an ecologist and evolutionary biologist from Princeton University, said the new study reminded her of her own research from last year, though the authors used a different approach. Tucci and her colleagues used genetic evidence to show that early modern humans mated with Denisovan populations on at least two different historical occasions, which means the lingering traces of Denisovan DNA embedded in the genomes of Asian populations living today are derived from at least two distinct Denisovan populations.


As for claiming the existence of some unknown human species, Tucci said the authors of the new study should refrain from conjecturing beyond the available evidence.

“While the study on the history of interaction between modern humans and archaic hominin contemporaries remains fascinating, I would be very cautious in making claims of admixture [genetic mixing] from an ‘unknown’ extinct archaic hominin population or speculate about the ‘discovery’ of a new hominin species,” Tucci told Gizmodo. “As the authors noted, the exact relationship of this extinct hominin population with the other known archaic populations (i.e. Neanderthals and Denisovans) is not completely disentangled—meaning that what the authors call an ‘unknown hominin’ could likely be a Denisovan group. More work—and more fossils—are needed to elucidate the history of archaic hominin admixture.”

No doubt, more evidence is needed. At the very least, however, these latest studies should encourage archeologists to be on the lookout for more clues pointing to the existence of the hypothesized species.


As a final word, the use of AI in archaeogenetics is a very cool development, and likely a sign of innovative research methods to come.

[Nature Communications]

Editor’s note: This article has been updated to change references to “hominid” to “hominin.” While the term hominid encompasses all great apes (including humans) and their ancestors, scientists now prefer the term hominin to refer specifically to humans and their ancestors. Smithsonian has a great article about the evolving vocabulary. Thanks to reader Scott Hatfield for pointing out this change in usage.