CRCNS US-Israel Research Proposal: Relating anatomy, multi-regional neural dynamics, and behavior during decision-making

Understanding how the brain integrates information in the service of decision-making is a fundamental question in neuroscience. The goal of this project is to build a theory of decision neuroscience that accounts for how the brain's intricate wiring gives rise to the neural activity behind a choice. This project builds that theory from existing data from very large numbers of neurons, collected while animals make decisions, together with detailed maps of the connections between them. From these it builds computer models showing how a circuit's structure shapes the choice itself: why a decision comes now rather than later, and which parts of it are a matter of chance rather than set by the brain's wiring. This work matters beyond understanding the brain itself, and may be useful for developing new artificial intelligence. The overall goal of the project is to identify the anatomically constrained circuit motif underlying decision-making. The major challenge of building such models is that brain-wide recordings find signatures of an upcoming choice across multiple regions in the brain. To overcome this challenge, the researchers propose to develop a new generation of Recurrent Neural Network models (RNNs) that reconstruct neural activity across multiple brain regions and behavioral variables at high resolution. The plan is to then build a generative model based on neural connectivity and develop an anatomically constrained multi-regional model of decision-making. The goal is then to use this model to identify the core connected brain regions in the model by perturbing them computationally to find which regions and pathways alter behavior when silenced. The overall result is a model constrained jointly by neural activity, anatomy, and behavior: its units are regions, cell types, and projections, and its dynamics reduce to a few variables that can be understood mathematically. The framework thus offers a general way to turn brain-wide recordings into mechanistic hypotheses, which can be used to guide future innovation. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.