Low-dimensional controllability of brain networks

Remy Ben Messaoud,Vincent Le Du,Brigitte Charlotte Kaufmann,Baptiste Couvy-Duchesne,Lara Migliaccio,Paolo Bartolomeo,Mario Chavez,Fabrizio De Vico Fallani

2023-11-28

Abstract:Network controllability is a powerful tool to study causal relationships in complex systems and identify the driver nodes for steering the network dynamics into desired states. However, due to ill-posed conditions, results become unreliable when the number of drivers becomes too small compared to the network size. This is a very common situation, particularly in real-world applications, where the possibility to access multiple nodes at the same time is limited by technological constraints, such as in the human brain. Although targeting smaller network parts might improve accuracy, challenges may remain for extremely unbalanced situations, when for example there is one single driver. To address this problem, we developed a mathematical framework that combines concepts from spectral graph theory and modern network science. Instead of controlling the original network dynamics, we aimed to control its low-dimensional embedding into the topological space derived from the network Laplacian. By performing extensive simulations on synthetic networks, we showed that a relatively low number of projected components is enough to improve the overall control accuracy, notably when dealing with very few drivers. Based on these findings, we introduced alternative low-dimensional controllability metrics and used them to identify the main driver areas of the human connectome obtained from N=6134 healthy individuals in the UK-biobank cohort. Results revealed previously unappreciated influential regions compared to standard approaches, enabled to draw control maps between distinct specialized large-scale brain systems, and yielded an anatomically-based understanding of hemispheric functional lateralization. Taken together, our results offered a theoretically-grounded solution to deal with network controllability in real-life applications and provided insights into the causal interactions of the human brain.

Neurons and Cognition,Quantitative Methods

What problem does this paper attempt to address?

This paper attempts to address the limitations of network controllability in practical applications, especially in the human brain network. Specifically, the paper focuses on how to improve the control precision when the number of driver nodes is much smaller than the network scale. Due to technical limitations, such as the limited ability to access multiple nodes simultaneously in the human brain, this leads to the unreliability of control results. In addition, the paper also explores how to improve network controllability through low - dimensional embedding, especially for extremely unbalanced situations, such as the situation when there is only one driver node. To this end, the author has developed a mathematical framework that combines spectral graph theory and output controllability, aiming to improve the overall control precision by controlling the low - dimensional embedding of the network, especially when dealing with very few driver nodes. Through this framework, the paper not only improves the control precision of synthetic networks, but also successfully identifies the main driving regions in the human connectome and provides a map of causal interactions between different large - scale brain systems, thus providing a new perspective for understanding brain functions.