Efficient scale-bridging methodologies for multi-scale modeling of the nervous system
Abstract
Despite decades of extensive research, simultaneously modeling multiple scales in complex biological systems such as the brain, from, low-level mechanisms (e.g., molecular interactions at the nanoscopic scale), to high-level observables (e.g., cognitive performance,) still constitutes a remarkable challenge. Yet, bridging this gap promises to shed much needed light on how the brain accomplishes,the prowesses it is capable of. The PI and his collaborators aim to develop cross-scale linking methodologies, thereby enabling a no,vel construct truly capable of encompassing multiple levels of complexity and capable of addressing the dichotomy that exists betw,een mechanisms at the microscopic scale, and consequences (whether positive or negative) observable at the organism scale.One of the, key challenges for multi-scale modeling is the large computer hardware requirement for performing simulations. Accurate mechanistic, models at the microscopic scale, which allow for the effects of molecular interactions to be simulated, are mathematically complex., Each cellular model can then have thousands of these microscopic mechanistic models which result in computationally expensive simul,ations. Cross-scale linking methodologies convert the mechanistic models into functional/statistical models using machine learning.,The functional models preserve the input-output properties of the mechanistic models but drastically reduce the computational comple,xity. The end result is a multi-scale model that can be (i) simulated using fewer computational resources and (ii) simulated for the, long periods of time that are necessary to observe effects at the behavioral/organism level.This research, focused more specificall,y on the hippocampus and cognition, may yield a better understanding of cognition, providing insights on the causes of potential dys,functions and how to alleviate them, or identify novel approaches to enhance it. This effort may also result in numerous high-impact, applications in fields as diverse as artificial intelligence (e.g., to improve generalized machine learning), or therapeutics ident,ification and development (e.g., antidote to nerve gas exposure), which are directly pertinent to the Armys mission. Yet, this effo,rt relies on the availability of extensive computing power to support (i) the simulations of multi-scale models, and (ii) the calibr,ation and optimization of efficient abstraction methodologies and data-driven models. Consequently, this DURIP application aims to s,ecure the acquisition of high-performance computing nodes (both CPU- and GPU-based) that will support and greatly accelerate our res,earch efforts dedicated to furthering ongoing Army-funded (ARO), and Army-related research projects (NIH, NSF). Notably, the enhance,d computational capabilities will also vastly improve training experiences for undergraduates, graduates, and postdoctoral students.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 01, 2022
- Source ID
- N000142212307
Entities
People
- Jean Bouteiller
Organizations
- Office of Naval Research
- United States Navy
- University of Southern California