MSU Scientists Teach Neural Network to Predict Molecular Properties More Accurately
Researchers at Lomonosov Moscow State University’s Artificial Intelligence Center have upgraded a neural network designed to predict molecular properties, introducing a method that analyzes a broader range of parameters and delivers more accurate results.
Today, scientists developing new drugs and advanced materials increasingly rely on machine-learning algorithms that analyze molecular structures and predict their physicochemical properties. Earlier approaches examined molecules only at the level of individual atoms. The Moscow-based researchers moved beyond that limitation: their neural network simultaneously analyzes both atoms and larger structural fragments known as functional groups.
In computational experiments, the new model demonstrated higher accuracy than traditional algorithms. According to the authors, the method could accelerate the search for chemical compounds with specific desired characteristics. The findings were published in the Journal of Chemical Information and Modeling.
