Percepta: Handling compounds outside the model’s applicability domain

Description
Percepta property predictions are most reliable for structures that fall within the applicability domain of each model. For unusual or very large structures, or those with rare functional groups, predictions may be accompanied by warnings or lower reliability scores.

Solution

• Run the desired property prediction in Percepta (e.g., logP, solubility, pKa, toxicity).
• After the prediction completes, check:
  • Any warnings about the structure being outside or at the edge of the applicability domain.
  • Reliability indicators (e.g., “Reliability” or “Confidence” scores, if provided).
• If the structure is flagged as outside the applicability domain:
  • Treat the numerical prediction as rough guidance, not a confirmed value.
  • Avoid using the prediction for critical decisions without supporting experimental data.
• Where possible:
  • Compare the structure to chemically similar compounds that are known to be within the model’s domain.
  • Use experimental data or targeted experiments to confirm or refine predicted values.