Abstract

In this talk, I will present an approach for interpreting ML models and will illustrate it through the example of the Molecular Transformer, the state-of-the-art model for reaction prediction. I will outline a framework to attribute predicted reaction outcomes both to specific parts of reactants, and to reactions in the training set. Furthermore, I will demonstrate how to retrieve evidence for predicted reaction outcomes, and understand counterintuitive predictions by scrutinising the data. Additionally, I will point out ”Clever Hans” predictions where the correct prediction is reached for the wrong reason due to dataset bias. Finally I will illustrate how the reported accuracy of models can be much higher than it is in reality due to not appropriate train-test splitting. For further details see: https://chemrxiv.org/articles/preprint/Quantitative_Interpretation_Explains_Machine_Learning_Models_for_Chemical_Reaction_Prediction_and_Uncovers_Bias/13061402