Yes. Sarah Nexus is a statistical system and will identify interactions that are sufficiently represented in the training set if they can form part of a larger fragment. It is always possible for Sarah Nexus to miss interactions if there is little evidence of this interaction in the training set or if contiguous fragments that capture the interacting fragments can’t be identified. The latter case could be where the scaffold holding the two interacting fragments is changed. Fragments that have some ‘global’ deactivating power (as is often seen with charged groups) will be recognised and will be learnt by the model.
We would always be delighted to hear of examples when Sarah Nexus or Derek Nexus gets a prediction wrong as this helps us improve our predictive performance as chemical space of interest to members changes.