The human muscle-type nicotinic acetylcholine receptor (nAChR) is an important pharmacological target whose activation involves rapid transition through intermediate pre-open states. Recent structural studies reveal that its two agonist-binding sites undergo independent and asynchronous conformational changes, creating an opportunity to identify ligands that preferentially engage one site over the other. Here, we applied a fragment-based screening workflow that combines a multi-objective machine-learning model with active learning to prioritize compounds from a fragment database. The model evaluates several binding-related features simultaneously and uses each screening round to guide the selection of new candidates. This approach enriched the candidate pool for fragments predicted to favor site-specific binding, and may provide a route toward ligands that probe, and potentially modulate, the asynchronous activation mechanism of the receptor.