Computational psychiatry: The next generation of clinical treatments

Leo Malchin | November 4th, 2024

It’s no surprise that the brain, the most complex organ in the body, gives rise to disorders that have long eluded enduring solutions. The emerging field of computational psychiatry is an ongoing attempt to demystify the medical community’s view of mental illness.

Unlike ailments of the body, our conceptions of psychological disorders morph in striking unison with shifts in our culture. The Diagnostic and Statistical Manual of Mental Disorders (DSM), which is meant to establish a common understanding of mental illnesses, widely reflects the state of psychiatry as a field. The DSM is updated every 10 to 15 years and gradually shifts in tandem with societal trends (as an example, the DSM listed “sexual orientation disturbance” as a pathology subject to treatment until 1980). 

It’s unlikely that such shifts will be impervious to today’s emphasis on big data and AI, and one can only assume that future clinicians will draw upon computational approaches to psychiatry as the basis for conceptualizing and treating mental illnesses.

Here emerges the field of computational psychiatry: a marriage between modern computational neuroscience techniques and the conventional — yet mostly archaic — model of psychiatry. Today, diagnoses depend on subjective assessments from clinicians, while patients must endure costly trials and errors before settling on an effective therapy. The goal of computational psychiatry is to bypass these inefficiencies and deliver precise treatments on an individual basis with data-driven solutions. 

Dr. Mark Wallace, a professor in Vanderbilt’s Department of Psychology whose lab leverages many of today’s state-of-the-art neuroscience techniques, sees great potential for disruption in the status quo of psychiatry.

“Psychiatry has been mired,” Wallace said. “Right now, the standard is to come up with a subjective diagnosis and then throw drugs at people, hoping that they work. Obviously, that helps some people, but we’re realizing that it’s not the right approach.”

Wallace and leaders in the emerging field of computational psychiatry like Michael Frank at Brown University categorize the discipline into two streams: data-driven and theory-driven. 

Data-driven approaches in computational psychiatry are those that use large, often multimodal datasets to build algorithms capable of efficiently clustering symptoms, reliably diagnosing disease, and accurately predicting treatment outcomes. The advent of machine learning and accessible neuroimaging and genome sequencing techniques have helped to enable such an approach. Ultimately, diagnoses produced by flexible algorithms that weigh objective parameters could be far more accurate than today’s standard of subjective evaluations by clinicians.

The theory-driven approach is more rooted in basic neuroscience research and seeks to understand the mechanisms behind disorders. Theory-driven computational psychiatrists look to gain a thorough understanding of how physiological differences map onto cognitive deficits, which may ultimately result in psychopathology. With this approach, specific disorders like autism or schizophrenia are defined in terms of overarching concepts like predictive coding. 

“The approaches that are ultimately going to win out are the ones that are both data-driven and mechanistic,” Wallace said. “I mentioned these two approaches as separate, but what we want to do is blend them together. You want to have a good algorithm, but you also don’t want to treat people as a black box.”

Wallace’s research focuses on multisensory processing rather than psychiatry, but a graduate of his lab, Dr. Albert Powers, dedicates his research at Yale University to understanding psychosis from a computational perspective. Powers eloquently frames schizophrenic auditory hallucinations in terms of a theory-driven computational approach. 

According to Powers, activity in brain circuits involving the superior temporal gyrus and excess dopamine in the striatum are biomarkers of both people with schizophrenia and people who are more susceptible to hallucination-inducing laboratory paradigms. These data identify highly specific neural signatures of hallucinations that may help researchers develop and prescribe more targeted drugs to those experiencing psychosis. This work is an example of modern neuroscience perspectives and advanced imaging techniques informing more accurate models of psychiatric disorders. Research in this domain produces specific characterizations of disorders, allowing for specific treatments.

Though research hubs in computational psychiatry are primarily clustered at institutions like Brown, Yale, and University College London, some clinical investigators at Vanderbilt, particularly of the younger generation, are beginning to shift toward a more computational approach.

Vanderbilt Assistant Professor of Psychology Antonia Kaczkurkin uses machine learning algorithms on large multimodal datasets to help restructure our current classifications of psychiatric disorders, but also to test and fine tune the algorithms themselves. 

Kaczkurkin is also involved in ENIGMA Anxiety Consortium, which pools neuroimaging data sourced from around the world to develop and train computational models of anxiety disorders. This communal effort to source large datasets allows them to gain statistically powerful insights on niche populations that are otherwise difficult to study. Complex computational models require lots of input in order to output valid predictions, so efforts like these to curate robust datasets are critical.

For a predictive model to work, it needs a functional algorithm and good, unbiased data. With the boom of AI in industry and academia, computer scientists are sure to produce feasible algorithms on which to base computational psychiatric models. Therefore, the future of computational psychiatry as a field depends largely on the availability of this kind of high-volume, high-quality data. 

“This is how all of medicine is going to evolve,” Wallace said. “We need to have all of these modes of information about people to know how to treat people. With the electronic medical record, which is becoming more and more conventionalized, there’s this push to have records in one common format. These kinds of things are seemingly trivial, but they are really big barriers to the development of good models.”

References

Bruin, W.B., Zhutovsky, P., van Wingen, G.A. et al. Brain-based classification of youth with anxiety disorders: transdiagnostic examinations within the ENIGMA-Anxiety database using machine learning. Nat. Mental Health 2, 104–118 (2024). https://doi.org/10.1038/s44220-023-00173-2

Corlett, P. R., Horga, G., Fletcher, P. C., Alderson-Day, B., Schmack, K., & Powers, A. R., 3rd (2019). Hallucinations and Strong Priors. Trends in cognitive sciences, 23(2), 114–127. https://doi.org/10.1016/j.tics.2018.12.001

Friston, K. (2022). Computational psychiatry: from synapses to sentience. Molecular Psychiatry. 28, 1-13. 10.1038/s41380-022-01743-z. 

Huys, Q.J., Maia, T.V., & Frank, M.J. (2016). Computational psychiatry as a bridge from neuroscience to clinical applications. Nature Neuroscience, 19, 404-413.

Drescher J. (2015). Out of DSM: Depathologizing Homosexuality. Behavioral sciences (Basel, Switzerland), 5(4), 565–575. https://doi.org/10.3390/bs5040565