Search for: All records

Abstract Objective. Improvements in recording technology for multi-region simultaneous recordings enable the study of interactions among distinct brain regions. However, a major computational challenge in studying cross-regional, or cross-population dynamics in general, is that the cross-population dynamics can be confounded or masked by within-population dynamics. Approach. Here, we propose cross-population prioritized linear dynamical modeling (CroP-LDM) to tackle this challenge. CroP-LDM learns the cross-population dynamics in terms of a set of latent states using a prioritized learning approach, such that they are not confounded by within-population dynamics. Further, CroP-LDM can infer the latent states both causally in time using only past neural activity and non-causally in time, unlike some prior dynamic methods whose inference is non-causal. Results. First, through comparisons with various LDM methods, we show that the prioritized learning objective in CroP-LDM is key for accurate learning of cross-population dynamics. Second, using multi-regional bilateral motor and premotor cortical recording during a naturalistic movement task, we demonstrate that CroP-LDM better learns cross-population dynamics compared to recent static and dynamic methods, even when using a low dimensionality. Finally, we demonstrate how CroP-LDM can quantify dominant interaction pathways across brain regions in an interpretable manner. Significance. Overall, these results show that our approach can be a useful framework for addressing challenges associated with modeling dynamics across brain regions. 

Neural dynamics can reflect intrinsic dynamics or dynamic inputs, such as sensory inputs or inputs from other brain regions. To avoid misinterpreting temporally structured inputs as intrinsic dynamics, dynamical models of neural activity should account for measured inputs. However, incorporating measured inputs remains elusive in joint dynamical modeling of neural-behavioral data, which is important for studying neural computations of behavior. We first show how training dynamical models of neural activity while considering behavior but not input or input but not behavior may lead to misinterpretations. We then develop an analytical learning method for linear dynamical models that simultaneously accounts for neural activity, behavior, and measured inputs. The method provides the capability to prioritize the learning of intrinsic behaviorally relevant neural dynamics and dissociate them from both other intrinsic dynamics and measured input dynamics. In data from a simulated brain with fixed intrinsic dynamics that performs different tasks, the method correctly finds the same intrinsic dynamics regardless of the task while other methods can be influenced by the task. In neural datasets from three subjects performing two different motor tasks with task instruction sensory inputs, the method reveals low-dimensional intrinsic neural dynamics that are missed by other methods and are more predictive of behavior and/or neural activity. The method also uniquely finds that the intrinsic behaviorally relevant neural dynamics are largely similar across the different subjects and tasks, whereas the overall neural dynamics are not. These input-driven dynamical models of neural-behavioral data can uncover intrinsic dynamics that may otherwise be missed.