Modern psychiatry is at a crossroads, as it attempts to balance neurological analysis with psychological assessment. Computational neuroscience offers a new lens through which to view such thorny issues as diagnosis, treatment, and integration with neurobiology. In this volume, psychiatrists and theoretical and computational neuroscientists consider the potential of computational approaches to psychiatric issues.

This unique collaboration yields surprising results, innovative synergies, and novel open questions. The contributors consider mechanisms of psychiatric disorders, the use of computation and imaging to model psychiatric disorders, ways that computation can inform psychiatric nosology, and specific applications of the computational approach.

---

This Forum is supported by the Deutsche Forschungsgemeinschaft

The German Research Foundation

Computational Psychiatry

What Can Theoretical Neuroscience and Psychiatry Teach Each Other?

Peter W. Kalivas and Martin P. Paulus, Chairs

Program Advisory Committee

Huda Akil, Joshua A. Gordon, Julia R. Lupp, John P. O'Doherty, Daniel S. Pine, A. David Redish and Klaas Enno Stephan

Background

We start from the idea that computational psychiatry entails a reciprocal interaction between theoretical/computational neuroscience and psychiatry. We believe that these two disciplines are mutually informative and have much to learn from each other.

We define theoretical neuroscience as a formal description of mechanisms that underlie measured behavioral and brain processes. Computational approaches are a key component of the formalisms that underlie theoretical neuroscience. These formal descriptions can provide new perspectives about brain-behavior relationships and enable predictions that can be used to guide experimental design and interpretation.  The hypothesis guiding this meeting is that these formal descriptions will prove to be useful in psychiatry, by informing classification, outcome-prediction, and therapeutics.

Goal

To discuss how new computational perspectives might be used to broaden our mechanistic understanding of psychiatric dysfunction and improve identification and treatment of psychiatric disorders.

---

This Forum is supported by the Deutsche Forschungsgemeinschaft

The German Research Foundation

Justification

A computational perspective asks different questions about patients than traditional clinical perspectives. These new questions have the potential to guide diagnosis and treatment by getting at fundamental psychological and neural processes, which may cut across symptomatic and diagnostic boundaries. These theoretical perspectives are based on the idea that brain mechanisms can be understood in terms of computational processes that function to transform information to produce behavior. In particular, this framework suggests that one of the major questions to ask is: “What is different about how this patient processes information about the world?”

For example, computational neuroscientists working on decision making have now garnered considerable evidence that suggests there are several action-selection systems that can drive choice behavior, each of which processes information about the world differently. These systems depend on different brain structures; damage to one structure can drive a patient to be more likely to use one or another system to make decisions. It is useful to consider how this computational perspective can guide clinical and translational studies in new directions unavailable without it.

However, these models and theories have not yet shaped critical thinking in psychiatry. It is our contention that part of the problem is a lack of communication between the computational groups and psychiatrists. Psychiatrists tend not to be trained in the computational perspective, and computational models tend not to be constructed to directly address the concerns of psychiatrists or patients. This Forum brings these two groups together to initiate a dialog geared at formulating how new computational breakthroughs can begin to have an impact on clinical practice. Four discussion groups have been constructed to approach this goal. The foci of the discussion groups will cut across computational neuroscience and psychiatry lines. To support an evolving dialog, each group will consist of both experts from psychiatry and experts in computational, theoretical, and basic neuroscience.

---

This Forum is supported by the Deutsche Forschungsgemeinschaft

The German Research Foundation

Group 1: How can computational approaches help identify fundamental mechanisms underlying psychiatric disorders?

• How might computational perspectives parse psychiatrically relevant behaviors into fundamental mechanisms?
• How could these fundamental mechanisms illuminate molecular, circuit, and system levels of analysis in psychiatry?
• How can a computational understanding identify biological markers and guide development of measures and assays that delineate dysfunction?

---

Group 2: Modeling Psychiatric Disorders Realistically: Incorporating complexity and heterogeneity into computational models

• What are the complexities within psychiatric syndromes that require a novel perspective?
• How can computational perspectives fully capture this complexity?
• What are the characteristics that would make a computational account successful in translation to psychiatry?
• Should we focus on symptoms or syndromes? How can theories and models move beyond single mechanisms?

---

Group 3: How can computational approaches inform diagnostic nosology?

• How do current nosologies fail? (prognosis? treatment? underlying mechanisms?) How might computational approaches address these failures?
• Are current frameworks (such as DSM or RDoC) a good starting point? Do computational perspectives provide a better place from which to start?
• What might a computational nosology look like?

---

Group 4: Candidate examples for a computational approach to address practical problems in psychiatry

• What are the characteristics of an application that makes it both feasible and relevant?
• What are the obstacles (computational, technological, practical) that could hinder success?
• Are there any examples where we are close to making a direct connection between the computational perspective and clinical practice? Are there any examples where the computational models have been able to capture a surprising aspect of the clinical phenomenon?
• How would this change a patient’s experience?

  ---

  This Forum is supported by the Deutsche Forschungsgemeinschaft

  The German Research Foundation