Inside Chemistry’s Graph Network Models — A Map of Reaction Formulas, Without Even Training on Reactions!
Recently, a very cool discovery has been made! Decision-making frameworks of graph neural network models in chemistry conceals a map of reaction formulas. Just like in natural language processing, “King” - “Man” + “Woman” = “Queen". Here it is the language of chemical reactions “reduced” - “H2” = “oxidized”.
Replicating the Hidden Knowledge Inside Graph Neural Networks— Reveals How GNNs are Transferable!
Building on previous research, we now possess the insights necessary to replicate the underlying knowledge within graph neural networks. This breakthrough demonstrates how these models can transfer learning to predict a wide range of chemical properties and accurately map reaction formulas!
One Model to Rule Them All: Transfer Learning From Graph Neural Networks
Is the hidden knowledge contained in graph neural networks enough for learning complete chemical behavior? Check out how we test this on many chemical properties such as electron occupancy, NMR, pKa, and solubility, all starting from the hidden knowledge contained inside graph models.
A Global Chemical Model Concealed Inside Graph Neural Networks
Graph Neural Networks are the some of the most sophisticated probabilistic AI models in chemistry, but how do they model chemical properties and behaviors? Dive in to see how we tap into the hidden decision-making framework of these complex AI models.
