Model Applicability

Metrics to assess confidences at every step of model building and predictions to help in the decisions such as the following are available in Sarchitect:

• Ascribe confidence to property predictions
• Applicability of models to chemical series
• Model expansion or model localization
• Selecting suitable algorithms and parameters

There are three Prediction Confidence metrics in Sarchitect:

• Descriptor Space Similarity
• MACCS keys based Tanimoto Similarity
• Algorithm Prediction Confidence

In addition, an easy-to-interpret composite confidence metric based on above metrics is available.

There are two methods to assess the applicability domain of the models vis-à-vis compounds of interest.

• Comparing Chemical Space of models against the prediction set.
• Fragment or Sub-structure searches