Anatomy-driven AI for translational neuroscience

Psychiatric illnesses derail lives. They are the biggest cause of disability in the UK, costing over £100 billion per year. Despite stunning advances in neuroscience, over the last 35 years treatments have not significantly improved. This unrealised potential may be, in part, due to sub-optimal selection of brain circuits and species for translational studies. By anchoring artificial intelligence (AI) to multiscale neuroanatomy, I aim to accelerate and  improve the selection process, bringing precision to the translational pipeline and so contributing to effective discovery of treatments for psychiatric illnesses.
State-of-the-art AI-inspired models of the brain lack anatomical details, making them unsuitable to tackle questions related to specific brain circuits and species like: “Which monkey brain circuits should we target to understand the biological mechanisms of hallucinations in schizophrenia?” or “If we uncover a mechanism for memory deficits in mice, is it likely to hold in humans?”. Answering such questions requires tools to translate emerging results between human subjects and animals, in terms of both neuroanatomy and brain-wide dynamic activity.
To this end, I will introduce two new computational approaches (SpeciesTransformer and Cortically-Embedded Recurrent Neural Networks – CERNNs). These models will integrate brain-wide multi-scale neuroanatomy data with AI, leveraging the burgeoning availability of high-quality cortex-wide gene expression data across species, the emergence of transformers in AI and my recent breakthroughs integrating multi-scale neuroanatomy and neural network models of cognition.
Objectives

Develop SpeciesTransformer to align cortical anatomy across species based on gene expression and multi-scale anatomy, enabling optimal selection of animal brain circuits for translational studies.
Develop the CERNN modelling framework to simulate cortex-wide neural dynamics during cognitive tasks based on neuroanatomy.
Simulate cognitive networks across species & compare with cross-species data, to enable evidence-based species selection for translational neuropsychiatry studies.

I have convened leading international collaborators, selected methodologies, and designed a research strategy that will allow us to identify optimal brain circuits and species for investigating cognitive dysfunction in psychiatry. Throughout the fellowship, an advisory group of world-leaders in translational neuropsychiatry from industry and academia will guide the work to maximise impact.
As a result of this project, neuroscientists and psychiatry researchers in academia and industry will benefit from improved translational mapping. Experimental and computational neuroscientists will avail of the openly-shared aligned anatomical data and model code to plan, analyse and interpret experiments. Over time, pharmaceutical companies, society and patients affected by mental illness will benefit through focusing resources on the experiments that are most likely to translate to human treatments.
Ultimately, this work will contribute to realising the promise of modern experimental neuroscience, generating more effective understanding and treatments for human psychiatric illness.