AD	Applicability Domain
QSAR/QSPR	Quantitative Structure–Activity/Property Relationship
OECD	Organization of Economic Co-operation and Development
QRPR model	Quantitative Reaction–Property Relationships model
ML-dependent AD	Applicability domain definition approaches that are tied to the machine learning method. It means that the determination of whether an object belongs to the model’s applicability domain and the prediction of property value for a given object is made by the machine-learning approach.
Universal AD	Applicability domain definition approaches that are not tied to the machine learning method. In this case, predicted property is given by Random Forest Regressor and the object’s belonging to the model’s applicability domain is determined in another way.
Leverage	Universal AD definition approach which based on the distance from the test sample objects to the centre of the training set distribution.
Lev_cv	Modification of Leverage approach. Leverage does not have internal hyperparameters, Lev_cv has one—the optimal threshold that needs to be adjusted.
Z-1NN	AD is defined on the basis of distance from a test set object to the most similar training set object(s). One nearest neighbour is considered.
Z-1NN_cv	Modification of Z-1NN_cv approach. Z-1NN doesn’t have internal hyperparameters, Z-1NN_cv has one—the optimal threshold that needs to be adjusted.
1-SVM	One-Class Support Vector Machine
2CC	Two-Class Classification method
RFClf	Random Forest Classifier
BB	Bounding Box
FC	Fragment Control
RTC_cv	Reaction Type Control. It needs to add «cv» since it is necessary to select the hyperparameter of the method
RTC*	Reaction Type Control with the first neighborhood
GPR	Gaussian Process Regression
RF	Random Forest Regressor
GPR-model	If structure-reactivity modelling is performed using Gaussian Process Regression
GPR-AD	The variance in prediction by Gaussian Process as AD measure
RFR_VAR	The variance in the ensemble of predictions as AD measure
Weak consensus	If most methods suggest that the reaction falls in AD, then it is predicted as belonging to AD
Strict consensus	The reaction belongs to AD if all the methods suggest that it falls in AD
OZ	“optimistic zero model”
PZ	“pessimistic zero model”
“Perfect AD model”	Exactly predicts inliers (reactions with prediction error less than 3xRMSE on cross-validation) and outliers (the rest).
RMSE	Root-mean-square error
R2	Coefficient of determination
∆R2_AD	It is the difference between the coefficient of determination for property prediction (R2) for objects within AD and corresponding value for all datapoints.
OIR	It is the difference between prediction RMSE for objects outside AD (denoted as RMSEOUT) and within AD (denoted as RMSEIN).
OD	Outliers Detection
AUC_AD	Area Under the ROC Curve as ranking criterion for AD detection
CV	Cross-validation
SN2	Bimolecular nucleophilic substitution
DA	Diels–Alder reactions
E2	Bimolecular elimination
CGR	Condensed Graph of Reaction