AI Speeds Catalyst Development for the Chemical Industry
Novosibirsk scientists have developed an algorithm that predicts where cations will be located in zeolites. The technology could accelerate the development of materials essential to oil refining and gas separation.
Researchers at the Boreskov Institute of Catalysis of the Siberian Branch of the Russian Academy of Sciences (SB RAS) have developed an algorithm that enables the targeted design of porous materials with specific properties. It replaces years of empirical trial and error with computational methods, significantly reducing costs and shortening the time it takes to bring advanced products to market.
The researchers developed an original approach. First, they simulated particle behavior across 4,000 different zeolite compositions. Using the resulting data, they built a kind of “dictionary” that translates a material’s chemical formula into numerical descriptors a computer can process. They then trained a neural network on this dataset. Rather than repeatedly calculating complex physical interactions from scratch, the network recognizes familiar structural patterns and produces results almost instantly. Its main advantage is versatility: the model works with a broad range of cations.
Researchers can now screen out unsuitable candidates on a computer, without spending time and money on physical laboratory testing.
Scientific Software Targets the Oil Refining Market
The Novosibirsk researchers’ development comes as demand grows for domestically produced catalysts. Since 2019, for example, Gazprom Neft PJSC has been building a state-of-the-art catalyst plant at its Omsk Refinery, designed to produce 21,000 metric tons annually. Meanwhile, in 2022, Rosneft Oil Company PJSC launched industrial production of a hydrocracking catalyst at a plant in Bashkortostan.
Yet high overall localization rates – 76% by the end of 2023, for example – can mask shortages in specific high-tech product categories, particularly those manufactured domestically. This gap is driving demand for Russian scientific software that can accelerate the design of new materials. Siberian researchers are addressing that need.
A Global Shift in Chemical Research
The adoption of neural networks in materials science is part of a global trend. In 2023, researchers at Southern Federal University applied machine learning algorithms to analyze the internal structure of zeolites using spectral data. A year later, ITMO University and chemists from Novosibirsk used machine learning to optimize catalyst parameters for hydrogen production. In 2025, the Boreskov Institute of Catalysis upgraded its vapor-phase crystallization method, significantly increasing the yield of zeolite nanocrystals. The 2026 algorithm builds on these advances, shifting the focus from physical synthesis toward intelligent data analysis.
Toward Homegrown Algorithms and Independent Databases
Over the medium term, the Novosibirsk-developed tool is slated for active deployment at research centers operated by chemical industry groups. This could allow them to replace foreign software with domestically developed alternatives.
The algorithm’s open-source code creates opportunities for commercialization. Russian organizations could offer foreign laboratories software tools for selecting sorbents, while also exporting physical catalysts designed with the Siberian neural network.
Building proprietary zeolite databases strengthens Russia’s technological independence in catalyst design and reduces the oil and gas industry's reliance on foreign software.
