Dual Support for Statistical Alerts and Expert Rules
CASE Ultra version 188.8.131.52 and above will support both Statistical Alert based models and Expert Rule based predictions. We have released Expert Rule Based System for predicting bacterial mutagenicity. This is in line with ICH M7 recommendation to test a chemical with two QSAR methodologies, i.e. statistical and expert rule based.
Statistical models are built with an automated learning process
During model building, CASE Ultra mines alerts (fragments of chemical structures) that are statistically related to the activity being modeled. A CASE Ultra model is therefore a collection of relevant alerts with a detailed account of their statistical relationship with the activity. In addition, a number of physicochemical descriptors (logP, water solubility, charges, E-states, surface descriptors, etc.) are employed to build local QSARs for each alert.
Predictions are based on alerts found in the test chemical
When a chemical is tested with a CASE Ultra model, a search is made through all the alerts of the model to see which ones are present in the test chemical. The final prediction is based on the identified alerts and their relationship with the model’s end point.
Test chemicals are checked for out-of-domain criteria
If no positive alerts are identified in a test chemical and part of its structure is not covered by the chemical space of the model being used, the test chemical is flagged as “out of domain” for the model. This is a good way of checking if the test chemical falls within the applicability domain of the model.