This webinar gives insight into recent development of ChemAxon's Calculators and Predictors towards trainable models and new ADMET models. Trainable models give the opportunity to use input data including experimental values and chemical structures to build models automatically using machine learning techniques. The pipeline connects descriptor generation, model building and validation and delivers a model ready to be used. New ADMET models refers to our plans to extend predictions with additional end-points. The talk introduces the first version of the hERG pAct regression model, as the first element in the ADMET group.