Distinguishing and integrating features of sensory input is essential to human survival and no less paramount in music perception and cognition. Yet, little is known about training-induced plasticity of neural mechanisms for auditory feature integration. This study aimed to contrast the two hypotheses that musical expertise leads to more independent (i.e. segregated) or more dependent (i.e. integrated) processing of acoustic features presented in contexts with high or low ecological validity with respect to music. To this end, mismatch negativity (MMNm) was recorded with magnetoencephalography (MEG) from 25 musicians and 25 non-musicians, exposed to sounds presented in interleaved blocks of a musical multi-feature paradigm and a classical oddball control paradigm. In addition to single deviants differing in pitch (P), intensity (I), or perceived location (L), double and triple deviants were included differing in combinations of these features (i.e. PI, IL, LP, PIL). Consistent with previous work, neural processing overlap was assessed in terms of MMNm additivity. Specifically, empirical MMNms obtained with double and triple deviants were compared to modelled MMNms corresponding to the sum of MMNms obtained with the constituent single deviants. Significantly greater MMNm under-additivity was observed in musicians compared to non-musicians, specifically for pitch-related deviants in the musical paradigm (i.e. PI, LP, and, marginally, PIL). Conversely, expertise effects were absent from the classical oddball control paradigm which used identical sounds. This novel finding supports the dependent processing hypothesis suggesting that musicians recruit overlapping neural resources facilitating more holistic representations of domain-relevant stimuli. These specialised refinements in predictive processing may enable musicians to optimally capitalise on some of the characteristic, complex variations in acoustic structure upon which music is based.