Scientists have developed a new artificial intelligence (AI) model that can predict how molecules move and change their shape over time much faster than older methods allow. This could help develop new medicines faster. Creating a new medicine often takes long and is costly. Thousands of possible molecules must be checked through repeated studies.
Molecular dynamics is used to study these movements. These methods calculate the forces between atoms and move each atom forward in very small time steps, billions of which are required. This makes the simulations very demanding on computers.
How the new AI model works
The new model learns the general rules of how molecules behave from many examples of short simulations. It does not calculate every tiny step one by one. Instead machine learning allows it to speed up each part of the process and to generate likely future arrangements of atoms directly.
The model proved more than ten thousand times faster than conventional simulations while still producing good results. The scientists tested it on more than twelve thousand five hundred organic molecules containing carbon nitrogen hydrogen and oxygen as well as over one thousand short peptides. The model observed movements over tens of nanoseconds and then predicted behavior over periods a thousand times longer.
A key advantage is that the model can handle molecules it never encountered during training because it learned broad patterns rather than memorizing individual cases. The predictions were checked against standard numerical calculations and matched well. Scientists describe the advance as the ability to jump between scenes in a molecular movie instead of watching every frame in order. In laboratory work the model could support measurements of specific molecular properties such as how readily a molecule dissolves in a solution or passes through a cell membrane. The study, published in Science Advances, was carried out by researchers at Chalmers University of Technology and the University of Gothenburg with one author also affiliated with AstraZeneca. The approach remains under development for more complex realistic conditions but it indicates how artificial intelligence can make early drug testing faster and more accurate while supporting broader understanding of molecular behavior in disease.