Digital Brain Seminar

Abstract: Uncertainty-aware inference from brain recordings is essential for identifying the underlying causal mechanisms of brain (dys)function, thereby advancing precision medicine, eg., through Virtual Brain Twins [1].

In this talk, I will first overview the state-of-the-art probabilistic machine learning approaches for inference, including likelihood-based methods (such as MCMC sampling) and likelihood-free methods (aka simulation-based inference), with the challenges involved in generalization. I will then present our recently developed tools, the flexible Virtual Brain Inference [2] and automatic Dynamic Causal Modeling [3], which are available in the cloud through EBRAINS services.

I will highlight the benefits of incorporating prior knowledge, reparameterization, and inference diagnostics for reliable causal inference. The performance of these methods will be demonstrated in epilepsy, multiple sclerosis, focal intervention, healthy aging, and social facilitation [4]. Finally, I will move beyond model-based inference and discuss learning brain dynamics directly from detailed simulations using deep neural networks. Overall, these results contribute to advances in precision medicine by enhancing the predictive power of Virtual Brain Twins.

---

[1] Hashemi et al. "Principles and operation of virtual brain twins." IEEE Reviews in Biomedical Engineering (2025).
[2] Ziaeemehr, et al. "Virtual Brain Inference (VBI), a flexible and integrative toolkit for efficient probabilistic inference on whole-brain models." Elife 14 (2025).
[3] Baldy et al. "Dynamic causal modelling in probabilistic programming languages." Journal of the Royal Society Interface 22.227 (2025).
[4] Esmaeili, et al. "Probabilistic inference of social presence across brain scales reveals enhanced synaptic efficacy." Communications Biology 8.1 (2025): 1608.

Abstract: The common neural mechanisms underlying the reduction of consciousness during sleep and anesthesia remain unclear. Previous studies have examined changes in network structure by only using recordings with limited spatial resolution, which has hindered the investigation of the critical spatial scales for the reduction of consciousness. To address this issue, we recorded calcium signals from approximately 10,000 neurons across multiple cortical regions in awake, sleeping, and anesthetized mice and compared network structure at different spatial scales by leveraging single-cell resolution and wide-field two-photon microscopy.

At the single-cell scale, both sleep and anesthesia exhibit higher network modularity than an awake state, indicating a segregated network, but modules are spatially intermixed in all three states. In contrast, at the mesoscale, there are no consistent differences in modularity between states, and modules are spatially localized. Our multi-scale analysis challenges the traditional view of network segregation during unconsciousness and indicates a scale-dependent network organization.