Russian Scientists Build AI That Searches DNA for Clues to Treat Cancer and Alzheimer’s

What These “Knots” in Our Genes Are and Why They Matter

Scientists from the AI Center at the National Research University Higher School of Economics have introduced DeepGQ, a model that predicts with high accuracy the location of key structures in human DNA known as G-quadruplexes. These structures influence whether a cell becomes a neuron or a liver cell, and how it may malfunction and turn cancerous. The new tool lays the groundwork for targeted drugs against some of the most severe diseases, many of which still lack effective treatment options.

Human DNA is not just a linear code made up of four letters, A, T, G, and C. To fit inside the cell nucleus and function properly, it folds into complex three-dimensional shapes. One such structural element is the G-quadruplex, or GQ. It can be visualized as a tight knot that forms in DNA regions rich in the base guanine.

This knot can physically block access to certain genes or, in other cases, promote their activation. It acts as a molecular switch. The specific combination of these switches in different tissues, such as the brain, liver, or heart, determines a cell’s identity and drives it to perform a specialized function.

For a long time, studying these structures was a costly and technically demanding experimental task. Researchers could examine only isolated DNA fragments and were unable to quickly build a comprehensive map of G-quadruplexes across the entire genome of a specific tissue. DeepGQ addresses this limitation using deep learning.

How DeepGQ Works and What Makes It Different

DeepGQ is an AI system that analyzes DNA sequences and predicts where G-quadruplexes are most likely to form. Its key distinction from existing tools is tissue specificity.

“Imagine you have a road map. A generic map shows only major highways,” says Maria Poptsova, Director of the Center for Biomedical Research and Technologies at the HSE AI Institute. “Our model is a detailed city map. For the brain, it reveals one network of GQ ‘streets’, while for the liver it shows a completely different one.”

Instead of a single universal model, the researchers trained 14 separate versions of the algorithm for different human tissue types. The system analyzes DNA in both directions, improving predictive accuracy, and incorporates epigenetic context, the molecular environment surrounding a gene. The result is not an averaged approximation but a tissue-specific map that closely reflects biological reality.

From Research to Clinical Practice

Accurate mapping of G-quadruplexes has direct medical relevance and could reshape approaches to treating some of the most challenging diseases.

Many severe conditions, including glioblastoma, an aggressive brain cancer, as well as Alzheimer’s and Parkinson’s diseases, are linked to deep disruptions in cellular programming. Cells lose their identity. G-quadruplexes that regulate this identity become attractive therapeutic targets. If researchers can determine which specific “knot” sustains uncontrolled cancer cell division, they could design a drug molecule that selectively disrupts it.

DeepGQ gives researchers access to a high-precision tool for predicting G-quadruplexes. Instead of costly experimental studies, any laboratory can take patient sample data and, using DeepGQ, obtain an accurate map of the most likely targets for validation

Artem Bashkatov

Junior Researcher, Center for Biomedical Research and Technologies, Institute of AI and Digital Sciences, HSE

The developers envision the next step as a “DeepGQ-Patient” framework. A physician could take a tumor sample, sequence its DNA, and use the algorithm to build a personalized map of active G-quadruplexes within that tumor. This would enable the selection or design of therapies tailored to the molecular profile of an individual cancer, potentially improving treatment effectiveness.

The approach also represents a breakthrough for rare disease research. The methodology allows large-scale, high-precision studies of how G-quadruplexes contribute to genetic and neurodegenerative disorders, accelerating the search for underlying mechanisms and actionable targets.

Why It Matters for Russia and Beyond

Publication of the results in the international journal Scientific Reports signals recognition of the strength of Russian research in AI for life sciences. It demonstrates the ability to develop competitive, world-class research tools.

For Russia, the project strengthens bioinformatics sovereignty. Advanced in-house genome analysis algorithms reduce dependence on foreign software in a strategically important field.

Globally, DeepGQ offers the scientific community a powerful and precise method that can already be applied to fundamental research in genetics. Over the longer term, it supports a future in which cancer and age-related neurodegenerative diseases can be treated with greater precision, lower toxicity, and therapies tailored to individual patients. Russian scientists and developers have shown they are playing a leading role in shaping that future.