A recurrent neural network-based framework for nonlinearly modeling behaviorally relevant neural dynamics

A key goal of several neuroscience studies is to understand and model how the dynamics of distinct populations of neurons give rise to specific human and animal behaviors. Many existing methods for exploring the link between neural activity and behavior rely on the analysis of static images and brain scans, as opposed to dynamically changing neural activity over time.

Dynamic models, mathematical or computer approaches to describe the evolution of a system over time constitute an interesting alternative to these methods. Most dynamic models introduced in the past were linear, meaning they assumed that changes in neuronal activity would follow a simple structure.

Although linear models tend to be easier to implement and interpret, they often fail to accurately capture complex neural dynamics. This has motivated some neuroscientists and computer scientists to develop other dynamic models capable of describing different types of linearity and nonlinear dynamics.

Researchers from the University of Southern California and the University of Pennsylvania recently introduced a new nonlinear dynamic modeling framework based on recurrent neural networks (RNN) that addresses some of the limitations of dynamic models introduced in the past for neuroscience research. This new framework, described in an article published in Natural neurosciencecan be used to model both relevant behavioral dynamics and other neural dynamics, but it decouples the two and prioritizes behavior-related dynamics.

“Understanding the dynamic transformation of neural activity into behavior requires new capabilities to nonlinearly model, dissociate, and prioritize behaviorally relevant neural dynamics and test hypotheses about the origin of nonlinearity,” they said. write Omid G. Sani, Bijan Pesaran and Maryam M. Shanechi in their article. . “We present Priority Dissociative Dynamics Analysis (DPAD), a nonlinear dynamic modeling approach that enables these capabilities with a multi-section neural network architecture and training approach.”

The researchers trained their RNN-based model using a four-step optimization algorithm. This algorithm allows the model to prioritize learning relevant behavioral latent states, while also learning any remaining neural dynamics.

To demonstrate the potential of their nonlinear dynamic modeling framework, the researchers applied it to five distinct problems in neuroscience. They specifically used it to analyze and model neuronal dynamics in datasets containing recordings of neuronal activity in the brains of non-human primates as they performed different tasks.

“By analyzing cortical spiking and local field potential activity in four movement tasks, we demonstrate five use cases,” Sani, Pesaran, and Shanechi wrote. “DPAD enabled more accurate neuro-behavioral prediction. It identified nonlinear dynamic transformations of local field potentials that were more predictive of behavior than traditional power features. Additionally, DPAD achieved reduction in neuronal dimensionality nonlinear prediction of behavior It allowed testing of hypotheses regarding nonlinearities in neuron-behavioral transformation, revealing that in our data sets, nonlinearities could be largely isolated from the mapping of latent cortical dynamics. behavior.

Results from initial testing by this team of researchers suggest that their model could be a valuable tool for neuroscience research, as it could help test hypotheses about how dynamic, non-neural dynamics relate to specific behaviors. . Notably, their model proved applicable to the study of behaviors that are continuous (i.e. monitored continuously for a given time), sampled intermittently (i.e. recorded at different times) and categorical (i.e. falling into distinct categories).

As part of their study, the researchers primarily demonstrated the use of their approach to model the transformation of primate neuronal activity into behavior. However, it could also be used to model other brain dynamics, such as the shared and distinct dynamics of different brain regions or neuronal dynamics caused by sensory stimuli.

© 2024 Science X Network