Model architecture for the AI method that predicts toxicity of chemicals
Photographer: Transformers enable accurate prediction of acute and chronic chemical toxicity in aquatic organisms
A representation of the molecule's structure is used as input to a pretrained transformer, which interprets the molecular structure. The transformer creates a so-called “vector embedding” – a numerical representation of the toxicity of the structure. That is then used as input to a deep neural network (DNN), together with information about the type of toxic effect you want to assess and the exposure duration. The output of the neural network is the predicted molecule concentration that causes the requested effect.
