Figure 15From: Feature combination networks for the interpretation of statistical machine learning models: application to Ames mutagenicityExample of a difference in models learning and interpretations elucidation. A model may learn that a pattern of aromatic nitro is activating. However, we are not able to generate a fragment describing this. As a result we would identify the activity as being caused by the aromatic nitro and the attached ring. The abstracted pattern is highlighted in blue on the ACTIVATING fragments.Back to article page