Functional processes in the human brain are shaped through complex interactions and correlations (relational links) between different brain regions. In neuroscience, the role of large-scale neural activities and inter-regional functional networks in cognitive processes is widely investigated; however, which of the correlations underlying the structure of these networks are most critical for predicting brain activity and how many connections are required for accurate modelling have not yet been fully elucidated. In the existing literature, the question of whether all connections between brain regions or only a specific subset determines the overall state of the system remains unanswered. In this context, the extent to which brain activity can be simplified or compressed in accordance with the principles of information theory is of importance for understanding neural data processing mechanisms.
In the study conducted by Weaver and colleagues, the objective was to quantify the compressibility of the human brain and to identify the most important correlations required to predict large-scale neural activity. Rather than assuming that the strongest connections are automatically the most important, the study aimed to determine optimal correlation networks with the highest informational content. The researchers tested whether a sparse correlation network (a structure containing a small number of connections) that minimizes uncertainty in brain activity would be sufficient to explain the system as a whole. Through this approach, it was intended to reveal weak but informative connections that may be overlooked by existing analytical methods and to increase efficiency in brain modelling.
Within the scope of the study, functional magnetic resonance imaging (fMRI) data from 99 healthy adults obtained from the Human Connectome Project database were analyzed. Cortical activities of the participants were examined both during resting state and during seven different cognitive tasks. In the analysis of the data, a computational method based on the minimax entropy principle derived from statistical physics and information theory was employed. With this method, correlations were added to the model one by one, and at each step, the connections that most reduced uncertainty were selected, thereby constructing optimal networks. The performance of the obtained optimal models was comparatively evaluated against networks formed from randomly selected connections and those constructed solely from the strongest correlations. The study demonstrated that the healthy human brain is highly compressible (C≈0.96) and that brain activity can be predicted with a very small number of correlations. The researchers note that examining how this property of compressibility changes in diseases such as Alzheimer’s, Parkinson’s, and schizophrenia constitutes a fundamental question. If these diseases disrupt the efficiency of information processing or the network structure of the brain, significant deviations in compressibility ratios may be observed in patients compared to healthy individuals. This suggests that compressibility (the C value) itself may be used as a diagnostic biomarker.
As a result of the analyses, it was determined that the human brain is highly compressible and that brain activity can be largely predicted using only a very small fraction of correlations. According to the findings, using only 1.4% of the existing correlations was sufficient to achieve a 50% reduction in uncertainty. Remarkably, the connections most critical for predicting activity were not always the strongest correlations. The strongest connections were observed to exhibit a tendency to cluster and to contain redundant information, whereas optimal networks displayed a sparser structure that formed bridges between different cognitive systems. These results indicate that brain activity is constrained by a simpler and sparser backbone of connections than previously assumed. Current neurological research often tends to focus on the strongest connections between brain regions; however, this study demonstrated that the connections most critical for predicting brain activity are not necessarily the strongest ones but rather weaker yet more information-rich (non-redundant) connections. This finding indicates that concentrating exclusively on robust signals may be inadequate for disease diagnosis. The researchers emphasize that identifying these specific interactions that dominate neural activity may provide significant implications for understanding neurological disorders and for developing targeted treatments.
Reference: Weaver, N. J., Faskowitz, J., Betzel, R. F., & Lynn, C. W. (2026). Quantifying the compressibility of the human brain. Proceedings of the National Academy of Sciences of the United States of America, 123(4), e2531115123. https://doi.org/10.1073/pnas.2531115123