Cognition, Anatomy and Neural Networks

A central goal is translation across species. By formalising what is conserved and what differs between mouse, marmoset, macaque and human cortex, we aim to predict when results from animal neuroscience will generalise to humans, and how to design experiments that are maximally informative for human brain and mental health research.

Our work sits at the intersection of cognitive computational neuroscience, brain mapping, artificial intelligence and psychiatry, with an emphasis on models that make testable predictions and support direct comparison to neuroscience data.

Projects and themes

Anatomy-constrained neural networks for cognition

We develop neural network models whose architecture and dynamics are explicitly constrained by cortical anatomy. These models are designed to capture cognitive computations while remaining directly comparable to brain-wide neuroimaging and physiology data, allowing us to ask not only what computation is being performed, but also where and why it is implemented in particular cortical systems.

Neuromodulation of cortex-wide distributed networks

How do common circuit motifs across the cortex generate large-scale dynamics underlying flexible functions? The answer may partly be due to gradients of receptors for neuromodulators such as dopamine, serotonin, acetylcholine and noradrenaline across the cortex. We combine dynamical systems modelling with large-scale brain organisation to explain how changing local parameters via neuromodulation can produce distinct regimes (e.g., stability, flexibility, gating) and predict their spatiotemporal dynamics across cortex.

Translational computational neuroscience

We contribute to and advocate for the emerging subfield of translational computational neuroscience. Our approach is to build computational models of cognition that explicitly integrate species-specific brain anatomy, and to test these models against rich cross-species datasets generated with our collaborators. By comparing model predictions across mouse, marmoset, macaque and human, we aim to identify when different species rely on shared versus distinct circuit and network mechanisms to solve similar cognitive problems.

Ultimately, our goal is to predict translatability: which preclinical findings are most likely to generalise to the human brain, and which are unlikely to do so without critical caveats. We also aim for this work to improve the ethical and scientific efficiency of neuroscience—by reducing studies in species or paradigms that are unlikely to translate, while identifying cases where questions can be addressed reliably in simpler models—and to accelerate preclinical research by focusing resources on the most appropriate species, circuits and measurements for a given translational target.

From synapses to brain-wide phenotypes and symptoms in mental health (stress and schizophrenia)

We develop mechanistic models to connect synaptic and circuit-level changes to cortex-wide dynamics, cognitive phenotypes, and ultimately symptoms. A major focus is understanding how neuromodulatory and microcircuit alterations associated with stress exposure and schizophrenia reshape distributed computations (e.g., working memory, perception and hallucinations), and why these changes affect particular brain networks and behaviour. By embedding these mechanisms within anatomically-grounded models, we aim to generate testable predictions about vulnerability, compensatory dynamics, and which interventions are most likely to normalise function across the relevant circuits and networks.

Principles of cortical organisation within and across species

We analyse organising principles of cortex by integrating multimodal brain data into modern AI/ML frameworks. We aim to learn shared and species-specific structure in cortical organisation across major species in cognitive neuroscience and psychiatry research. This work also directly informs our neural models of cognition.