AutoQSAR Output Properties
The table below lists the Maestro properties that are added when a prediction is made for a structure. The value of label in the property name is the label you set in the AutoQSAR Prediction text box.
| Property | Description |
|---|---|
| Pred label | Value of the predicted property. Reported for numeric models. |
| Pred label SD | Standard deviation in the predicted numeric property, evaluated over the models used for the prediction. If only one model is used, the value is 0. |
| Pred label Class | Class of the predicted property. Reported for categorical models. |
| Pred label Prob | Probability of the structure belonging to the predicted class, for categorical models. This is the average of the probabilities calculated by the models that predict the reported class. |
| Pred label Purity | Fraction of models that predict the reported class, for categorical models. If only one model is used, the value is 1. |
| Domain label Score | Score that indicates whether the structure is in the domain of applicability of the model. The score is based on fingerprint similarity and is normalized so that a value of 0 corresponds to the training set average, and the scale is in standard deviations. A score of 1 means one standard deviation from the training set average. The cutoff for structures that are outside the domain depends on the diversity of the structures in the model. The value is averaged if multiple models are used in the prediction. |
| Domain label Alert |
Flag for structures outside the domain of applicability. For a single model, this property is set to 1 only if the domain score is less than -2. If the domain score is positive, the compound has greater-than-average similarity to the training set, so the alert is not needed. The value is averaged if multiple models are used in the prediction, so a value of 0.8 means that the structure is outside the domain of applicability in 8 out of 10 models. |