Figure 6. Candidate compound selection and activity prediction.

In ( a), a SARM is shown that represents a highly continuous local SAR environment. In this case, the potency of a virtual compound can be predicted using the NBH-based approach. By contrast, ( b) shows a SARM representing a discontinuous local SAR. Activity cliff-forming compound pairs are highlighted in blue. Such SAR environments fall outside the applicability domain of NBH-based potency predictions. However, marked VCs represent promising candidates for compound design based on their proximity to activity cliffs. Both SARMs originate from a set of cannabinoid CB1 receptor ligands (compound structures are omitted for clarity).