Pharma has been using QSAR/QSPR and machine learning, along with various other statistical models for 50+ years. There have been repeated over-predictions as e

Pharma has been using QSAR/QSPR and machine learning, along with various other statistical models for 50+ years. There have been repeated over-predictions as each new technology has come down the line. Yet the principles, like in this article, hold - have a good training set, have a test set that isn't just an echo of the training set and then finish with a validation set that is sufficiently different to help expose edge failures if not outright failures. Remember Egan's rule of 1 - the number of times a modeler can fail before their predictions are ignored...