Bridging the gap between biological plausibility and function in models of prefrontal cortex (360G-Wellcome-215909_Z_19_Z)

Prefrontal cortex (PFC) plays a fundamental role when we make complex decisions. However, there are very few biologically plausible neural-network models that can explain PFC activity during decision making. Additionally, most studies focus on sensory information that does not change during trials and decisions must be made at specific times, unlike our more dynamic natural environment. We aim to combine experimental recordings and biologically plausible neural-network models to identify decision-making mechanisms in PFC for a large variety of tasks. Our models will be biologically plausible in the sense that they will consist of distinct excitatory and inhibitory populations, we will train networks using reward-based training, and model neural activity will not be unnecessarily complex and nonlinear, in line with experimental recordings. We will examine the role that nonlinear neural activity plays in typical decision-making tasks and we will contrast this with more naturalistic tasks where sensory stimuli can change and decision times may vary. We will also examine learning-induced changes in the models from reward-based training with new recordings of PFC while monkeys learn context-dependent decision-making tasks. Finally, we will study circuit mechanisms in both our models and the new experimental recordings that enable rapid learning of new similar tasks.