Skip to main content

Transforming the understanding
and treatment of mental illnesses.

Banner of Shelli Avenevoli, Ph.D.

Computational Neuroscience: Deciphering the Complex Brain

By

I wrote in my welcome message about my priorities. First, we need to fund excellent science. Second, we should support studies that will yield benefits on short, medium, and long-term timescales. I also have three particular areas of interest: neural circuits, computational and theoretical psychiatry, and suicide prevention. Here I will discuss computational and theoretical approaches to mental health research. These approaches can be applied across the entire NIMH portfolio, and have the potential to yield benefits in the short, medium, and long-term.

On an unusually hot day in Frankfurt in the summer of 2015, 44 scientists sat in a sweltering room as fans hummed in the background. Along with my co-organizer David Redish, a neuroscientist at the University of Minnesota, we had brought together a group comprised of theoretical neuroscientists, schooled in the challenging art of creating detailed computational models of the brain, and psychiatrists, experienced in the equally challenging art of diagnosing and treating individuals suffering from the most misunderstood of brain disorders. They had spent the previous four days divided into small groups, learning how to talk to each other while simultaneously wrestling with the most challenging issues our field faces.

Now they had come together for one culminating discussion. As one of the young neuroscientists tasked with reporting from her group tried to sum up a week’s worth of discussion, a member of one of the other groups interrupted, “I don’t understand. Why do you have the diagnoses alongside the behavioral data, as if diagnoses and behavior are both caused by the illnesses? Aren’t diagnoses the illnesses themselves? Don’t they in turn cause abnormal behavior?”

From the side of the room, a psychiatrist from the neuroscientist’s group answered the question: “No, no, the diagnoses aren’t illnesses, they are observations, made by clinicians and probabilistically caused by the underlying illness. Bayes’ theorem* lets us work backward from the diagnosis to determine the probability that the patient has the illness, based on the clinician’s observation.”

Stunned, we all fell silent, suddenly aware of the humming of the fans. Stunned, by the incisiveness of the point, but also by how the interaction between computational and clinical expertise had led to a new perspective and new way forward that likely would have been impossible from either field alone.

Indeed, that was the point of the gathering Dave and I put together, to get theoreticians and psychiatrists to talk to, and even better, to understand one another. We succeeded because each field has so much to offer the other.

Take first the perspective of the psychiatrist. The science of psychiatry is a study in contrasts. On the one hand, advances in genetics, basic neuroscience, neuroimaging, and big data approaches hold the promise to dramatically improve knowledge of and treatments for mental illnesses. On the other hand, progress has been frustratingly slow, leaving us with few if any biomarkers, indefinite and subjective diagnostic categories, and partially effective treatments.

Why the slow progress? A principal factor is the complexity of the brain and the complexity of how neural systems produce behavior. It has proven incredibly challenging to connect knowledge gained at genetic, circuit, systems, and behavioral levels. How do the genes that predispose to schizophrenia alter function in the circuits that govern the cognitive processes disturbed in patients with the disorder? And how might circuit dysfunction lead to changes seen in neuroimaging? How can we understand the incredible heterogeneity seen in patients with the disorder, even those who share the same genes.

Computational approaches have the potential to answer these questions. Computational approaches allow us to describe and test how complex high-level phenomena emerge from interactions at smaller scale levels. Computational models of neural circuits that take into account differences in genetic makeup can put into explicit mathematical terms testable hypotheses regarding how alterations in genes might affect circuit function. Similarly, computational models of circuit dysfunction can test how such dysfunction could create a progressive, chronic disorder by impacting neural development and plasticity, and how that dysfunction could be revealed in neuroimaging findings and manifest in behavior. Finally, computational approaches can help take advantage of large data sets, categorizing brain dysfunction in a way that has the potential to lead to better diagnoses and improved biomarkers. 

Meanwhile, from the perspective of the theoretical or computational neuroscientist, psychiatry presents a set of real-world, clinically relevant problems, characterized by a particularly complex set of heterogeneous phenomena that are continually changing and responding to feedback. The problems psychiatry faces—understanding brain function from gene to cell to circuit to behavior; charting disease-related changes longitudinally, through development, and in the face of plasticity and homeostatic mechanisms; clarifying and categorizing illnesses in the face of multiple factors influencing risk and heterogeneity in how they present —provide challenges to even the most creative theoretician. But they also offer opportunities to contribute to both understanding the brain and improving the lives of those who suffer from brain illnesses.

NIMH recognizes the key contributions computational and theoretical approaches can make to psychiatry, as well as the need for psychiatrists to provide their expertise and insights with regard to the real-life problems that individuals with mental illnesses face on a daily basis. For these reasons we will be working hard to encourage collaborative efforts to integrate theory and computation into both existing and novel research programs. First steps include defining priorities within this area to focus on, a process we have already started in conjunction with extramural and intramural scientists. We have also encouraged the submission of applications for supplements to existing grants, to add  or enhance  computational and theoretical components. These and other efforts will be built on keeping the dialog going between theoreticians and psychiatrists, with the aim of teaching us all how to speak the same language, so that we can accelerate knowledge and improve the lives of individuals with mental illnesses.

For a more in-depth description of this issue, see the introductory chapters of the book1 David and I edited that grew out of the forum mentioned here. These chapters can be found online . Please note that neither of us have a financial interest in the book.

  1. A.D. Redish and J. A. Gordon (2016) Computational Psychiatry: New Perspectives on Mental Illness. MIT Press.

*For those unfamiliar with Bayes' theorem, it is a mathematical way of determining the likelihood that a hypothesis is true given the evidence at hand. What, for example, is the likelihood that a person who has had a heart attack has a particular genetic factor that predisposes to high cholesterol? Or a problem with the heart’s pacemaker? Bayes’ theorem offers a process for approaching such questions; from the dialog between psychiatrists and computational neuroscientists at our meeting emerged the insight into how the same thinking might assist in refining psychiatric diagnoses.