1 Computational mechanisms underlying thalamic regulation of prefrontal signal-to-noise ratio in decision making Xiaohan Zhang 1 , Michael M. Halassa 2 and Zhe Sage Chen 1,3,4 * 1. Department of Psychiatry, New York University Grossman School of Medicine, New York, NY, USA 2. Department of Neuroscience, Tufts University School of Medicine, Boston, MA, USA 3. Neuroscience Institute, New York University Grossman School of Medicine, New York, NY, USA 4. Department of Biomedical Engineering, New York University Tandon School of Engineering, New York, NY, USA. * Email correspondence: zhe.chen@nyulangone.org (Z.S.C.) Abstract The mediodorsal (MD) thalamus is a critical partner for the prefrontal cortex (PFC) in cognitive flexibility. Animal experiments have shown that the MD enhances prefrontal signal-to-noise ratio (SNR) in decision making under uncertainty. However, the computational mechanisms of this cognitive process remain unclear. Here we use performance-optimized computational models to dissect these mechanisms. We find that the inclusion of an MD-like feedforward module increases robustness to sensory noise and enhances working memory maintenance in the recurrent PFC network performing a context-dependent decision-making task. Incorporating genetically identified thalamocortical pathways that regulate signal amplification and noise reduction further improves performance and replicates key neurophysiological findings of neuronal tuning. Our model reveals a key computational mechanism of context-invariant, cell-type specific regulation of sensory uncertainty in a task-phase specific manner. Additionally, it makes experimentally testable predictions that connect disrupted thalamocortical connectivity with classical theories of prefrontal excitation-inhibition (E/I) imbalance and dysfunctional inhibitory cell types. MAIN Successful execution of complex decision-making tasks requires identification and processing of multiple sources of uncertainty 1 . Task uncertainty may appear in the form of corrupted or incongruent sensory cues (sensory uncertainty) 2,3 , their mapping to onto internal or behavioral variables (mapping uncertainty) 1,4,5 , or their likelihood of resulting in reward (outcome uncertainty) 6 . Experiments across multiple species have sown that the MD thalamus is a critical partner for the PFC in resolving uncertainty in decision making 7-13 . Human neuroimaging studies have shown that MD activity tracks sensory uncertainty in a multi-attribute attention task 13 , as well as a categorization task 15 . This process generalizes to non-human animals; in mice performing a decision-making task, the MD tracks sensory uncertainty and outputs a signal that appears to enhance prefrontal SNR 5,16 . Optical manipulations support this notion and delineate their causal roles in cognitive control 4,17 . However, the computational mechanism by which the MD enhances prefrontal activity in decision making under uncertainty remain unclear. In addition, how the newly discovered cellular diversity in MD thalamus contributes to such computations is unexplored. Biologically-inspired computational modeling provides a powerful approach to probe important questions that yield mechanistic insight into neural circuits 18-25 . Development of computational