Abstract
Adaptive cognition relies on cooperation across anatomically distributed brain circuits. However, specialised neural systems are also in constant competition for limited processing resources. How does the brain's network architecture enable it to balance these cooperative and competitive tendencies? Here we use computational whole-brain modelling to examine the dynamical and computational relevance of cooperative and competitive interactions in the mammalian connectome. Across human, macaque, and mouse we show that the architecture of the models that most faithfully reproduce brain activity, consistently combines modular cooperative interactions with diffuse, long-range competitive interactions. The model with competitive interactions consistently outperforms the cooperative-only model, with excellent fit to both spatial and dynamical properties of the living brain, which were not explicitly optimised but rather emerge spontaneously. Competitive interactions in the effective connectivity produce greater levels of synergistic information and local-global hierarchy, and lead to superior computational capacity when used for neuromorphic computing. Altogether, this work provides a mechanistic link between network architecture, dynamical properties, and computation in the mammalian brain.
Introduction
Traditional whole-brain models often assume that long-range connections between brain regions are purely "cooperative" or excitatory. This assumption stems largely from the limitations of structural imaging tools like diffusion MRI, which primarily map positive anatomical weights. However, competition and inhibition are fundamental to neural circuits at the microscopic level. This paper demonstrates that incorporating macroscale competitive interactions is essential for capturing the rich, complex dynamics of the living brain, such as the anticorrelations observed in functional MRI data.
Methods: A Cross-Species Modeling Framework
The researchers employed computational whole-brain modeling using Hopf oscillators, where each brain region is modeled as an oscillator poised near a critical point between noise and stable rhythm. The core innovation is the transition from standard structural connectivity to Generative Effective Connectivity (GEC). Using a heuristic gradient algorithm, the authors iteratively updated connection weights to best fit empirical fMRI data across two conditions: a cooperative-only model (positive weights only) and a cooperative-plus-competitive model (allowing for negative weights).
The study utilized high-quality datasets for three mammalian species:
- Human: 100 unrelated subjects from the Human Connectome Project (HCP).
- Macaque: 19 sessions from the PRIME-DE database, combining diffusion MRI and tract-tracing.
- Mouse: 10 subjects using high-resolution axonal tract-tracing from the Allen Mouse Brain Atlas.
Results: The Necessity of Competition
Across all species, the models that included competitive interactions significantly outperformed cooperative-only models. In humans, the correlation between simulated and empirical brain activity more than doubled, jumping from r = 0.42 to r = 0.87. Similar dramatic improvements were seen in macaques and mice.
The best-fitting models revealed a specific organizational principle: modular cooperation (strong positive ties within local clusters) complemented by diffuse, long-range competition (weaker negative ties across the brain). This architecture acts as a stabilizer, preventing runaway synchrony and allowing for a healthy balance of integration and segregation. Furthermore, competitive models were significantly better at identifying individual "fingerprints," capturing the unique subject-specificity of brain dynamics.
Computational Capacity and Cognitive Matching
Beyond fitting data, the researchers tested the functional implications of these models. When used as a "reservoir" for neuromorphic memory tasks, competitive models consistently outperformed cooperative ones, suggesting that the "tug-of-war" between regions increases the brain's information-processing capacity. In humans, the competitive model also produced activity patterns that better matched 123 meta-analytic maps of cognitive functions (e.g., attention, memory) compared to standard models.
Discussion and Future Work
While this study proves that macroscale competition is mathematically necessary for realistic brain modeling, the exact biological mechanism remains unclear. It is yet to be determined if this competition arises from direct long-range inhibitory projections or via long-range excitation of local inhibitory interneurons. Future work integrating neurotransmitter maps or MR spectroscopy will be essential to distinguish between these possibilities and to explore how this balance shifts in clinical conditions like Alzheimer’s or stroke.
Summary updated