HSE Researchers Teach Neural Network to Distinguish Origins from Genetically Similar Populations

Researchers from the AI and Digital Science Institute, HSE Faculty of Computer Science, have proposed a new approach based on advanced machine learning techniques to determine a person’s genetic origin with high accuracy. This method uses graph neural networks, which make it possible to distinguish even very closely related populations.
Over the past 10–15 years, genetic analysis has become increasingly popular not only as a tool for medical diagnostics, but also as a means of ancestry research. DNA testing allows people to learn more about their ethnic background, identify the places where their ancestors lived, and determine the number of Neanderthal mutations in a person’s genome.
This has become possible thanks to the development of modern technologies—such as genotyping, data storage and processing systems, and machine learning—and the significant reduction in their cost. However, current testing methods are unable to differentiate between genetically similar populations that have lived in adjacent regions for extended periods.
Researchers from the AI and Digital Science Institute have developed a method for distinguishing between individuals from closely related populations. At the heart of this technology are graph neural networks, which do not rely on DNA sequences but instead use graphs to represent genetic links between individuals with shared genome segments. These shared segments indicate the degree of kinship between people, revealing how many generations back their common ancestors lived. The more overlaps there are, the closer their ancestral connection is. In the model, each person is represented by a vertex in the graph, and the strength of the connection between them is indicated by the edges in the graph.
The method was tested on data from various regions. The results were particularly insightful for the population of the East European Plain, as a large dataset had already been compiled there. The graph neural network was able to accurately determine the population affiliation of individuals from genetically similar ethnic groups.
Aleksei Shmelev
‘Existing methods of genetic analysis address a different task: they identify affiliation with large, isolated groups, such as determining whether someone has French, German, or English ancestry. Our method enables the analysis of closely related populations, which is particularly relevant for Russia, a country with a diverse ethnic background,’ said Aleksei Shmelev, one of the study's authors and Research Assistant at the HSE International Laboratory of Statistical and Computational Genomics, AI and Digital Science Institute.
In their future work, the researchers aim to train the neural network to predict the proportion of different populations within a genome.
They have named their development AncestryGNN, which stands for 'Neural Network-Based Prediction of Population Affiliation via Shared Genome Segments.’
Vladimir Shchur
As noted by Vladimir Shchur, Head of the International Laboratory of Statistical and Computational Genomics at the AI and Digital Science Institute, HSE University, the proposed method holds great potential for more accurate understanding of human history and can be applied in genealogy and anthropology research.
This research was supported by a grant from the Government of the Russian Federation as part of the federal program ‘Artificial Intelligence.’
See also:
Physicists Show How a Solvent Shapes the Response of Drug Molecules
Many drug molecules can switch between different three-dimensional shapes. Using a mathematical model of ibuprofen dissolved in supercritical CO₂, researchers from HSE University and the Institute of Solution Chemistry of the Russian Academy of Sciences have shown that the solvent can make these molecules appear to respond unusually slowly. The molecules themselves continue to change shape rapidly, but their surroundings rearrange much more slowly, and this lag shows up in the measurements. The study has been published in The Journal of Chemical Physics.
Researchers Develop New Method to Detect Anomalies in Data Centre Cooling Systems
Researchers from HSE University and Ural Federal University have developed an AI-based method for early detection of gradual refrigerant leaks and heat exchanger fouling in data centre cooling systems. By accounting for the physical connections between the components of the refrigeration circuit, the new model can identify anomalies at an early stage. The results have been published in IEEE Transactions on Industry Applications.
Researchers Use iFORA to Map the Future of Industry 4.0
HSE University presented the capabilities of iFORA, an intelligent big data analysis system, at Kazan Digital Week 2026. An overview of technological opportunities and potential threats associated with Industry 4.0 was delivered by Konstantin Vishnevskiy, Director of the Centre for Strategic Analysis and Big Data at HSE ISSEK.
Physicists Train AI to Optimise Particle Beams in Accelerators
Researchers from the AI and Digital Science Institute at the HSE Faculty of Computer Science and the Budker Institute of Nuclear Physics at the Siberian Branch of the Russian Academy of Sciences have developed RLABC (Reinforcement Learning for Accelerator Beamline Control), a framework that uses reinforcement learning to identify optimal settings for a particle accelerator. In a computer simulation, RLABC optimised the accelerator's parameters to achieve a particle transmission rate of 70%. The study has been published in Computer Physics Communications.
HSE Scientists Present Tensor Methods Research at International Conference in China
Researchers from the AI and Digital Science Institute at the HSE Faculty of Computer Science took part in the 3rd Conference–School on Tensor Methods in Mathematics and Artificial Intelligence Computing. The event was held in Shenzhen, China, in late October 2026. One of the institute's research projects received the Best Poster Award, placing among the conference's five best posters.
Flattened Terrain Models Underestimate Vehicle-to-Vehicle Communication Outages
Researchers at HSE MIEM have shown that widely used simulators of vehicle-to-vehicle communication significantly overestimate its reliability because they assume flat terrain. When terrain features, bridges, and tunnels are taken into account, the outage rate on a mountain serpentine road rises from 21% to 30%, on an overpass from 20% to 33%, and in a tunnel from 0% to 80%. The results have been published in Future Transportation.
Study Reveals Why Experienced Readers Sometimes Make Mistakes in Comprehension
Researchers at the HSE Centre for Language and Brain examined how people comprehend sentences in which the grammatical structure conveys a meaning that conflicts with assumptions based on their life experience. In an experiment, participants were asked to read sentences describing two characters and then identify which character had performed the action. The results showed that people with greater reading experience performed better when a character's actions matched their expectations but were more likely to make mistakes when the actions contradicted those expectations. The paper has been published in the Quarterly Journal of Experimental Psychology. The study was supported by a grant from the Russian Science Foundation.
Fall into ML 2026: From Fundamental Research to Discussions on the Future of AI
On October 23–24, HSE University hosted Fall into ML 2026, the fifth conference on machine learning and AI. Over the past five years, the event has grown from a small gathering of researchers into a flagship platform that brings together leading machine learning specialists, early-career researchers, students, and representatives of technology companies. The fifth edition marked a new stage in the development of the conference, with a programme that included discussions with international speakers, presentations of joint research projects, and tutorials on the latest trends in AI.
The Future of Humanoid Robots in the Home Explored at Foresight Session
On October 23–24, HSE University hosted the landmark fifth Fall into ML 2026 Conference. During the foresight session titled ‘A Humanoid in Your Kitchen: When Will It Happen and What Can Prevent It?’ researchers explored why humanoid robot assistants have not yet become part of our everyday lives, what needs to change for this to happen, and who will be held responsible when such assistants make mistakes.
‘A Reliable Bridge Between a Laboratory Prototype and a Working Service’
The HSE Faculty of Computer Science will complete the development of the third phase of the SmartMLOps system by the end of the year. This enterprise platform simplifies and accelerates the deployment of AI models in production environments. The work is being carried out as part of HSE University’s ‘Multi-Agent Platform of AI Solutions for Industry-Specific Tasks.’ Hadi Saleh, Head of the Unit for Applied Technological Solutions at the AI and Digital Science Institute of the HSE Faculty of Computer Science, explains what makes the new system unique and what challenges it will be able to address.


