Create a new Relative Reasoning to limit the number of degradation products in the prediction. You just need to set a name and the packages you want the object to be based on. If you want to use the default options suggested by us, simply click Submit, otherwise click Advanced Options.

Select advanced options for the Relative Reasoning creation. The threshold is used for the predicted probabilies and should be between 0.0 and 1.0, the fingerprinter calculates the features for each compound, and the Classifier type is the used multi-label classifier.

Selection of parameter values for the Applicability Domain process. Number of Neighbours refers to a requirement on the minimum number of compounds from the training dataset that has at least one triggered transformation rule that is common with the compound being analyzed. Reliability Threshold is a requirement on the average tanimoto distance to the set number of "nearest neighbours" (Number of neighbours with the smallest tanimoto distances). Local Compatibility Threshold is a requirement on the average F1 score determined from the number of nearest neighbours, using their respective precision and recall values computed from the agreement between their observed and triggered rules. You can learn more about it in our wiki!