Multi-scale modelling & digital twins for bipolar disorder
Three projects, spanning theory to application: two integrated theoretical models of why the bipolar brain destabilises, and an applied programme that turns either into a forecasting tool for individual patients — in service of a more personalised psychiatry.
Two theories, one shared target
Project 1 and Project 2 each test a different, independently falsifiable account of why the bipolar brain loses stability — one through homeostasis, one through criticality. Neither depends on the other, and neither is a strict prerequisite for Project 3. But both hand it a common currency: a recovery rate and a distance to a tipping point, the two quantities that Project 3's digital twins fit for each patient. A digital twin can be built from data alone — but it is stronger, and more interpretable, when grounded in either or both theories. If Projects 1 and 2 are refuted, Project 3 still delivers a validated forecasting tool, and the refutations are themselves a substantive result.
Homeostasis dysregulation as a core component of Bipolar Disorder: the HDBD multi-scale model
Homeostasis — the capacity of a system to restore equilibrium after a perturbation — is foundational throughout medicine, yet rarely applied formally to the brain. The HDBD model asks whether bipolar disorder reflects a reduced capacity to restore equilibrium after perturbation, at neural, circadian and behavioural scales at once.
We apply computational frameworks for homeostatic regulation to existing patient data spanning fMRI, EEG, actigraphy, pupillary response and ecological momentary assessment, comparing how well patients and controls return to baseline after everyday perturbations.
Full model: Houenou J. Homeostatic Dysregulation: A Multi-Level Integrated Model of Bipolar Disorder. PsyArXiv, 2025 — osf.io/preprints/psyarxiv/9yrma_v1
Brain criticality
Complex systems theory suggests the healthy brain operates near a "critical" point, maximising flexibility and information-processing efficiency. Moving away from that point should lower the threshold at which a small perturbation triggers a large, system-wide change.
Using existing EEG and fMRI data and the tools of dynamical systems theory, we test whether the bipolar brain sits further from criticality than the healthy brain, and whether that distance tracks clinical instability.
Digital twins
A digital twin, as we use the term, is an individually parameterised generative model of a patient's mood dynamics — fitted to that patient's own data and evaluated only on data it has not seen. Our twins are multi-layered, spanning neural networks to behaviour, and are constructible from data alone; the theoretical models above provide grounding rather than a prerequisite.
Mood is modelled as motion in a multistable landscape under stochastic forcing, combined with whole-brain simulations constrained by each participant's own connectome. Individual fits yield a recovery rate, a barrier height between mood states, and a distance to the next transition.