Researchers from Chalmers University of Technology and the University of Gothenburg have developed an AI model that accelerates molecular simulations by more than 10,000 times compared to conventional methods. This advancement, detailed in a study published in Science Advances on June 11 2026, aims to improve the efficiency of early-stage drug development by predicting how molecular structures evolve over time. Traditionally, molecular dynamics simulations calculate atomic forces at intervals of approximately one femtosecond (10⁻¹⁵ seconds), requiring billions of steps that are computationally intensive. The new model, named TITO (Transferable Implicit Transfer Operators), learns the underlying statistical rules governing molecular motion, allowing it to predict transitions over nanosecond scales without performing step-by-step numerical calculations.

The study tested the model on over 12,500 organic molecules, including carbon, nitrogen, hydrogen, and oxygen-based compounds, as well as more than a thousand short peptides. Unlike conventional methods that require granular tracking, TITO identifies patterns in molecular evolution, enabling researchers to predict properties and changes in systems the model has not previously encountered. According to lead author Juan Viguera Diez, an industrial doctoral student at AstraZeneca, the model functions by learning the physics of how a system behaves, providing insights into molecular shapes, speeds, and pathways. The team validated these AI-generated predictions using standard numerical algorithms to ensure consistency with physical laws.

Researchers believe this capability could significantly reduce the time and costs associated with testing thousands of potential drug candidates. By accurately simulating molecular behavior, the tool helps identify promising medicines more quickly. Although the current version has been tested on small molecular systems in simplified solvent models at specific temperatures, ongoing development is focused on scaling the technology for more complex and realistic biological environments. This effort is seen as an important step toward facilitating faster drug discovery and expanding the understanding of disease mechanisms through advanced computational predictive modeling.

Researchers from Chalmers University of Technology and the University of Gothenburg have developed an AI model that accelerates molecular simulations by more than 10,000 times compared to conventional methods. This advancement, detailed in a study published in Science Advances on June 11 2026, aims to improve the efficiency of early-stage drug development by predicting how molecular structures evolve over time. Traditionally, molecular dynamics simulations calculate atomic forces at intervals of approximately one femtosecond (10⁻¹⁵ seconds), requiring billions of steps that are computationally intensive. The new model, named TITO (Transferable Implicit Transfer Operators), learns the underlying statistical rules governing molecular motion, allowing it to predict transitions over nanosecond scales without performing step-by-step numerical calculations.

The study tested the model on over 12,500 organic molecules, including carbon, nitrogen, hydrogen, and oxygen-based compounds, as well as more than a thousand short peptides. Unlike conventional methods that require granular tracking, TITO identifies patterns in molecular evolution, enabling researchers to predict properties and changes in systems the model has not previously encountered. According to lead author Juan Viguera Diez, an industrial doctoral student at AstraZeneca, the model functions by learning the physics of how a system behaves, providing insights into molecular shapes, speeds, and pathways. The team validated these AI-generated predictions using standard numerical algorithms to ensure consistency with physical laws.

Researchers believe this capability could significantly reduce the time and costs associated with testing thousands of potential drug candidates. By accurately simulating molecular behavior, the tool helps identify promising medicines more quickly. Although the current version has been tested on small molecular systems in simplified solvent models at specific temperatures, ongoing development is focused on scaling the technology for more complex and realistic biological environments. This effort is seen as an important step toward facilitating faster drug discovery and expanding the understanding of disease mechanisms through advanced computational predictive modeling.