Fascinating new neuroscience model predicts intelligence by mapping the brain’s internal clocks

A new study suggests that the brain processes information with high efficiency by synchronizing the physical wiring of neural networks with the varying speeds of local brain activity. Published in Nature Communications, this research offers a mathematical framework that aligns the brain’s structural connections with the timing of its electrical pulses. The findings indicate that models accounting for these varied internal speeds can better predict individual cognitive abilities compared to traditional approaches.

To understand the study, it is necessary to first grasp the concept of the connectome. The human brain consists of billions of neurons connected by a dense web of white matter fibers. Neuroscientists refer to this comprehensive map of neural connections as the connectome. This physical structure serves as the highway system upon which brain activity travels. While the structure remains relatively static, the activity within it is dynamic and constantly changing.

Brain regions do not all operate at the same tempo. Some areas, such as those responsible for processing sight and sound, must react almost instantly to incoming stimuli. Other areas, particularly those involved in complex thought and decision-making, integrate information over longer periods. These characteristic speeds are known as intrinsic neural timescales.

Engineers and mathematicians often use a framework called Network Control Theory to study complex systems. This approach models how a system moves from one state to another based on its connectivity and inputs. Neuroscientists have adapted this theory to model how the brain switches between different patterns of activity. A persistent limitation in this field has been the assumption that all brain regions function at the same speed.

Standard models typically assign a uniform time constant to every node in the network. This simplification makes the mathematics easier to handle but fails to reflect biological reality. Jason Z. Kim from Cornell University and Linden Parkes from Rutgers University sought to correct this discrepancy. Along with their colleagues, they developed a new model that infers the specific timescale of each brain region based on its behavior.

The researchers hypothesized that a model allowing for variable timescales would provide a more accurate representation of brain function. They used data from the Human Connectome Project, which includes brain scans from hundreds of young adults. The team utilized functional magnetic resonance imaging to observe how brain activity patterns shift over time. They also used diffusion-weighted imaging to map the structural white matter connections for the same individuals.

The team designed an algorithm to “learn” the internal decay rates of different brain regions. In this context, the decay rate represents how quickly a burst of activity fades away in a specific area. A faster decay corresponds to a shorter timescale, while a slower decay indicates a longer window for processing information. The algorithm adjusted these rates until the model could accurately simulate the transition from one brain state to another.

One of the primary measures the researchers looked at was “control energy.” In control theory, energy represents the magnitude of the input required to drive a system from a starting point to a desired end point. A highly efficient system requires less control energy to achieve a transition. The researchers found that their optimized model required consistently less energy than the standard uniform model.

This reduction in energy suggests that the brain is naturally wired to leverage these diverse timescales. By aligning the speed of local processing with the global network structure, the brain minimizes the metabolic cost of thinking and reacting. The researchers validated this finding by comparing their results against random null models. They found that the energy savings were specific to the actual anatomy of the human brain.

The study also demonstrated that this optimization allows the brain to be controlled by fewer inputs. In the uniform model, a simulation might require inputs to almost every region to successfully guide the brain’s state. The optimized model achieved the same transitions by stimulating a much smaller subset of regions. This finding has potential implications for understanding how localized neural signals can influence global brain states.

To confirm that their mathematical values corresponded to biological reality, the authors compared their model’s timescales with gene expression maps. They utilized the Allen Human Brain Atlas, a detailed dataset showing which genes are active in different parts of the cortex. The researchers looked specifically at genes related to inhibitory interneurons, which are cells that regulate the timing of neural firing.

Two specific markers of inhibitory cells, somatostatin and parvalbumin, show distinct patterns across the brain. Parvalbumin-expressing cells are typically associated with fast signaling and sensory processing. Somatostatin-expressing cells are linked to slower regulatory processes. The researchers found a strong correlation between their model-based timescales and the density of these molecular markers.

Regions that the model identified as having fast timescales showed higher expression of genes associated with parvalbumin. Conversely, regions with slow timescales in the model were rich in genes related to somatostatin. This biological validation indicates that the mathematical optimization successfully captured the underlying cellular architecture of the cortex. The model derived these values solely from imaging data, without prior knowledge of the gene maps.

The team also examined whether these findings held true across different species. They applied the same modeling approach to high-resolution connectivity data from mice. The results mirrored those found in humans. The mouse model showed similar improvements in energy efficiency and exhibited the same correlations with inhibitory cell markers.

This cross-species consistency suggests that the alignment of structural connectivity and neural timescales is a fundamental principle of brain organization. Evolution appears to have conserved this efficient architecture. The findings imply that the coordination between macroscale wiring and microscale cellular properties is essential for mammalian brain function.

Beyond general biological principles, the researchers investigated whether their model could explain individual differences in humans. They fit their optimized model to the specific brain scans of each participant in the study. This generated a unique set of timescales for every individual. The team then checked how well these personalized models tracked with the participants’ actual brain activity during rest.

Participants whose intrinsic timescales were better aligned with their structural connections tended to transition more frequently between different brain states. This suggests that a well-tuned brain is more dynamic and flexible. The researchers then engaged in a predictive modeling exercise. They attempted to forecast participants’ scores on various cognitive tests based on the properties of their brain models.

The optimized model outperformed the standard uniform model in predicting cognitive behavior. Features derived from the variable-timescale model showed stronger associations with performance on tasks involving fluid intelligence and spatial orientation. This indicates that the subtle variations in how fast different brain regions operate are relevant for higher-order cognition.

The authors noted several caveats to their work. The study relied on magnetic resonance imaging, which has limitations in temporal resolution. Neural activity happens on the order of milliseconds, while the imaging data captures changes over seconds. Consequently, the model likely captures a smoothed approximation of the true neural dynamics.

Additionally, the structural maps used in the study cannot distinguish the direction of information flow along nerve fibers. The researchers had to assume bidirectional connections for the human data, which is a simplification of the actual biology. However, the successful replication in the mouse dataset, which used directed connectivity data, mitigates this concern to some degree.

Future research will likely focus on how these timescales change during development and aging. The brain undergoes massive structural reorganization during childhood and adolescence. Tracking how intrinsic timescales evolve alongside these structural changes could provide insights into the maturation of cognitive abilities.

There is also potential for applying this framework to the study of psychiatric and neurological disorders. Conditions such as schizophrenia and autism are often described as network disorders involving disruptions in brain connectivity. It is possible that these conditions also involve a mismatch between the brain’s physical wiring and its temporal processing speeds.

The study, “Inferring intrinsic neural timescales using optimal control theory,” was authored by Jason Z. Kim, Richard F. Betzel, Ahmad Beyh, Amber Howell, Amy Kuceyeski, Bart Larsen, Caio Seguin, Xi-Han Zhang, Avram Holmes and Linden Parkes.