Abstract
Expectation suppression (ES) refers to the attenuation of neural responses when sensory inputs align with prior predictions, reflecting a core principle of efficient neural coding. To investigate the circuit-level mechanisms underlying ES, we developed a biologically inspired computational model incorporating excitatory pyramidal cells (PCs) and three major types of inhibitory interneurons: parvalbumin-expressing (PV), somatostatin-expressing (SOM), and vasoactive intestinal peptide-expressing (VIP) neurons. The model included stimulus-selective tuning in both excitatory and inhibitory populations and implemented experience-dependent synaptic plasticity. Our results revealed that SOM neurons mediated feature-specific dendritic inhibition of PCs under expected conditions, selectively suppressing responses to predicted inputs. Conversely, VIP neurons facilitated disinhibition by inhibiting PV neurons, thereby enhancing PC responsiveness under unexpected conditions. These dynamic roles arise from their distinct connectivity profiles and synaptic weight adaptations following training. Crucially, stimulus selectivity in SOM and VIP neurons was essential for precise modulation: without it, inhibitory control became diffuse and inefficient. In contrast, PV neurons provided broad, non-selective inhibition across PC populations, serving as a stable baseline for prediction-dependent modulation. Together, these findings demonstrated how interneuron diversity, synaptic plasticity, and tuning specificity interacted to implement ES in a prediction-sensitive manner.