Deep Learning and Human Evolution
By Ramisa Rifa
The evolution and the origins of humans has been studied for centuries. Archaeological digs, ground penetrating radar, and the careful study of any and all DNA that scientists can salvage from the wreckages of our ancestors, has been used to build a portrait of the human experience, down to its very roots. However, as our understanding of modern and prehistoric DNA deepened, gaps in the evolution of humans were revealed. Ghost ancestors and sub-species were discovered with the help of AI technologies, and their DNA is still preserved in us today.
Neanderthals (Homo neanderthalensis) and Denisovans (Denisova hominins) are both extinct subspecies of archaic humans. Evidence suggests that anatomically modern humans interbred with them multiple times as they traveled in and out of Africa. As such, modern Africans have much less Neanderthal DNA compared to European and Asian populations, 0.5% and 1.7% to 1.8% respectively. Denisovan DNA makes up 4 to 6% of melanesian DNA, and is present in lower percentages in Southeast Asian and Pacific Islander populations. This means that a certain percentage of highly variable DNA, single poly-nucleotide polymorphism (SNP), can be from Neanderthals or Denisovans. These SNP’s are evidence of interbreeding, also called introgressions, that occurred in the past.
Scientists compared the chromosomes of two humans, and found more differences than one would expect in the DNA of individuals from the same species. The Neanderthal genome was sequenced in 2010 and later the Denisovan genome, which resulted in the discovery that these divergences were fragments of DNA that came from said ancestors. In their study, published in Nature Communications, Bertranpetit et al, found these introgressions and identified whether they came from Neanderthals or Denisovans, canceled out those fragments, and studied the remaining DNA. They discovered that something about the DNA was still highly divergent. When the body of a teenage girl was found in Siberia’s Denisova cave, it was determined that she was the hybrid of a Neanderthal mother and a Denisovan father. This hybrid species was exactly what scientists hypothesized that could be the originator of this ghost DNA. They attempted to find all these divergences and analyze the genetic combinations that could have produced them, with the help of deep learning algorithms.
Deep learning (DL) is a type of artificial neural network that is designed to work the way a mammalian brain would. These algorithms are able to “learn,” using patterns it detects in large swaths of data to find information or perform tasks. The statistical model of Approximate Bayesian Computation (ABC) which is commonly used in population genetics, was useful in its flexibility and ability to be used with estimated parameters in situations where there are infinite numbers of possible operations and models. ABC was paired with DL to identify single nucleotide polymorphisms in the distribution of allele frequencies across the population of ancient and modern DNA. The deep learning algorithm was able to compress this data and obtain patterns that distinguish among proposed models and specified values for parameters of demographic interest. The ABC-DL algorithm successfully differentiated among eight introgression models and suggested a third major introgression event happened that was mutual to both Asian and Oceania populations, along with the first introgression with neanderthals in the major out-of-Africa event of anatomically modern humans and the Denisovan introgression with Oceania populations.
In short, the researchers used the non-linear nature of ABC-DL algorithms to test eight of the most plausible models of human evolution. The algorithm simulated how human DNA and demographics may have evolved, accounting for the nature and structure of DNA, as well as the complex patterns of migration throughout Afro-Eurasia. Over weeks of computation, the researchers test how well each model reaches the contemporary human genetic composition. The most data-backed models supported that a ghost ancestor had introgressed with anatomically modern humans in Asia, called Xe. Data suggests that the Xe lineage had its own long, independent revolutionary history, and that it was either a sister lineage of Denisovans or was closely related to both Neanderthals and Denisovans, with more similarities to Denisovans.
Much is still not known about Xe, and the many other ghost ancestors that were likely to have contributed to our genomes, but the discovery of their presence is major for understanding the complexities of early human societies and human evolution as a whole. The importance of understanding our genetics fully is crucial to our society as a whole, whether in curing disease or for the sake of history. To understand human evolution is to connect with ourselves and our ancestors on a deeper level, to learn and grow from them.