Researchers at the Zelinsky Institute of Organic Chemistry of the Russian Academy of Sciences have developed a neural network that emulates the intuitive decision-making process of experienced chemists. This is reported by the
official website of the Ministry of Science and Higher Education of the Russian Federation.

For the first time, an artificial intelligence system has been trained not merely to analyse molecular structures but to replicate the way scientists think when designing new materials and compounds.

The researchers applied a machine learning technique known as Learning to Rank (LTR), commonly used in search engines and recommendation systems. Adapted for chemical science, LTR enables the neural network to learn from expert assessments rather than follow predefined rules. Instead of relying on fixed parameters, the algorithm was trained to understand what human experts consider important when assessing molecular complexity.

A unique dataset was compiled for this purpose, comprising around 300,000 molecules evaluated by 50 professional chemists. Scientists compared molecular structures in pairs, determining which were more complex based on expert judgement. These comparisons generated nearly 200,000 data points, allowing the neural network to learn to rank molecules with a level of accuracy closely resembling that of human specialists.

The result is a neural network capable of evaluating molecular complexity with expert-level precision. It marks the first time an AI system has successfully replicated the cognitive processes behind chemical intuition, rather than functioning solely within rigid mathematical frameworks. The model also enables the transformation of molecular complexity from a subjective concept into a measurable, analysable characteristic.

According to the developers, this breakthrough offers a powerful tool for scientific advancement in a variety of disciplines, including pharmaceuticals, materials science, and synthetic chemistry.

Photo:
iStock