Select Page

Computational Psychiatry: Is Mathematics the Cure for Madness?

Can a mental disorder be calculated? Can any disorder be put into numbers at all? It’s certainly hard to imagine, without understanding some of the major functions of the brain. The human brain performs functions which are of a computational nature. Therefore, some scientists claim, a computational approach is needed to understand disorders of the brain function better — particularly psychiatric ones, where biological disturbances are extremely subtle and hard to track. 

The young field of computational psychiatry attempts a new, improved understanding, prognosis, and treatment of mental illness. It’s an interdisciplinary field, which includes psychiatry, experimental and clinical psychology, neuroscience, machine learning, artificial intelligence and computational neuroscience. The combination of those results in two types of approaches: gathering data topically and attempting to build mathematical or computational models of the relevant neural, circuit, or cognitive processes.

Those who are new to the field are often intimidated by the need to gain understanding of both mental health and formal methods. The teaching programs rarely include everything at once. Is mathematics really necessary to treat patients? And which mental health symptoms should be the main objects for formal study?

The fairly new book Computational Psychiatry: A Primer, authored by Janine Simmons, Brice Cuthbert, Joshua Gordon and Michele Ferrante from the US National Institute of Mental Health, brings new insight into the field. Until now, when approaching computational psychiatry, students who were interested in the field needed to acquire expertise in various areas, but each of them separately: maths, machine learning, computational neuroscience, reinforcement learning, psychotherapy, psychiatry, neuroscience. It’s a long list and, rightfully, put students off. 

The beginning of Computational Psychiatry: A Primer has a broad introduction, which is followed by chapters with more details on the current state of computational understanding of schizophrenia, depression, anxiety, addiction and tic disorders. They give a focused and resourceful overview of the recent history of psychiatry.

Among the most important computational methods are highlighted: Lapicque’s integrate-and-fire model of neurons, Rall’s cable theory, Hodgkin and Huxley’s description of action potentials, Hebb’s plasticity rules, and Barlow’s information theoretical characterization of sensory adaptation to today’s models of reinforcement learning and neural networks. 

The book also provides a tour through the main theoretical approaches, identifying the key formalisms and outlining their applications; in-depth reviews of biophysically based neural network models, cognitive control, and reinforcement learning as applied to issues in mental health; explanations of how dynamical models enable cellular-level processes to be related to high-level phenomena, for instance, how alterations in receptor dynamics affect working memory. “The book has many strengths,” says Quentin J.M. Huys in a 2022 review of Computational Psychiatry, “and much to like. It should become a useful and approachable, hence important, introductory text to those interested in the field.”

It is expected that this field will likely substantially advance psychiatry in the near future. However, there is a downside to it. The data-driven approaches are limited in their ability to fully capture the complexities of interacting variables in and across multiple levels. On the other hand, theory-driven approaches are yet to be applied to clinical problems.

Even though the treatment outcomes appear to be very promising, the computational tools have a number of limitations too: they require substantial expertise of a trained user. Another major challenge is generating a fruitful exchange between clinicians, experimentalists, trialists and theorists. This, says Huys in the article Computational psychiatry as a bridge from neuroscience to clinical applications from March 2016, “might be helped by a stronger focus on establishing utility by actively pursuing computational approaches in clinical trials.”

Overall, there are many standard clinical and theoretical boundaries still, and their integration remains untested at large, but computational psychiatry opens up many new opportunities to gain insight into mental illness, and ultimately, promises better outcomes for patients.