CoChat

The promises and pitfalls of applying computational models to neurological and psychiatric disorders

Open access

The DOI matches its Crossref record; 66 of the 66 references it cites were found in a registry or an index; none of the 49 sources its claims cite is checked for support yet.

Abstract

Computational models have become an integral part of basic neuroscience and have facilitated some of the major advances in the field. More recently, such models have also been applied to the understanding of disruptions in brain function. In this review, using examples and a simple analogy, we discuss the potential for computational models to inform our understanding of brain function and dysfunction. We argue that they may provide, in unprecedented detail, an understanding of the neurobiological and mental basis of brain disorders and that such insights will be key to progress in diagnosis and treatment. However, there are also potential problems attending this approach. We highlight these and identify simple principles that should always govern the use of computational models in clinical neuroscience, noting especially the importance of a clear specification of a model’s purpose and of the mapping between mathematical concepts and reality. Computational models are applied increasingly to the study of brain function and dysfunction. Teufel & Fletcher highlight the promise of this approach, but also some of the problems that can arise from the misapplication of such models. Using a simple analogy, they identify key principles necessary for avoiding common pitfalls.

Overview

CoChat Summary

TL;DR

Teufel & Fletcher review the promises and pitfalls of applying computational models to neurological and psychiatric disorders, arguing such models can enforce precision, inspire new conceptualizations, and bridge explanatory levels, but can mislead when their purpose and mapping to reality are unclear. They propose two principles—specify the model’s purpose/level and explicitly map mathematical concepts to mental or neurobiological processes—plus simplicity, generality, and breakability.

Key findings

  • The paper identifies three main benefits of computational models in clinical neuroscience: (i) enforcing rigour and precision in formalizing conceptual models; (ii) inspiring useful new conceptualizations and synthesizing disparate evidence; and (iii) bridging levels of explanation from basic neurobiology to conscious experience of suffering.
  • Examples include the jumping-to-conclusions bias, where a formal model suggested noise in decision-making rather than sampling costs; Plaut and Behrmann’s neural network, which predicted face and word processing impairments in prosopagnosia and pure alexia that were subsequently confirmed empirically; and Bayesian/predictive coding accounts of hallucinations that can support profoundly different conceptualizations.
  • The authors propose two key principles: clearly specify the purpose and level of explanation of a model, and explicitly treat the mapping between mental/neurobiological processes and mathematical concepts.
  • Good computational models are described as often simple, general, and breakable; only models making specific, testable predictions can be broken, and failures should drive iterative updating.

Methods

  • Conceptual/review article; no primary empirical data, sample, or statistical analysis are reported.
  • Uses a running/fitness analogy and illustrative examples from computational psychiatry and neurology to discuss benefits and pitfalls.
  • Proposes principles and criteria for evaluating computational models rather than testing a specific model.

Limitations

  • The paper is a conceptual commentary/review, not an empirical study; no systematic review or primary data are reported (evident from the text).
  • The authors state that computational clinical neuroscience is heterogeneous and that models can be opaque, mutable, and inexact; they warn against stretching explanatory power beyond what models can unambiguously support.
  • The examples are illustrative and selective; the text does not provide a formal meta-analysis or exhaustive coverage (evident from the text).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1093/brain/aww209

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Graphical abstract Computational models are applied increasingly to the study of brain function and dysfunction. Teufel & Fletcher highlight the promise of this approach, but also some of the problems that can arise from the misapplication of such models. Using a simple analogy, they identify key principles necessary for avoiding common pitfalls.
Graphical abstract Computational models are applied increasingly to the study of brain function and dysfunction. Teufel & Fletcher highlight the promise of this approach, but also some of the problems that can arise from the misapplication of such models. Using a simple analogy, they identify key principles necessary for avoiding common pitfalls.CC BY · PubMed Central
Figure 1 The importance of specifying both the purpose of a model, and the mapping between the model components and aspects of the ‘real world’. Here, we illustrate the importance of two principles of modelling using the running analogy from Box 1 . It is tempting, when we have identified a reliable measure, one that seems to capture the essence of what we are trying to characterize and that has predictive value, to make it generally applicable. We may forget or ignore both the elements of the world that we are not including in our model and, furthermore, the ways in which the components that we are modelling may map to the real world. The figure illustrates how we might use ‘Running Economy’ (the volume of oxygen consumed at steady state running) as described in Box 1 as a predictor of race performance. Having the figures for oxygen consumption on two athletes allows a direct comparison of how they are likely to fare in competition. Indeed, even if the measurements were obtained at a range of speeds and on different terrain, the measures may be comparable. Running economy serves as a good model of running ability. But, as indicated in the figure, it only fulfils this purpose within predefined constraints. For example, it is standard to measure running economy at a pace that is below that at which a person is near their lactate threshold (the point at which oxygen consumption cannot keep up with demand) and therefore the model is inadequate if we wish to judge running ability for shorter distance sprints in which lactate levels are highly relevant ( Billat, 1996 ). Here the mapping between the model and reality has gone wrong because the model is applied for a setting where its assumptions are no longer relevant or valid. Unless we are explicit about our mapping between model components and reality and the working assumptions that justify it, our model will ultimately lead us astray. Such error lies not in the model but in how it is used. It should also be noted that running economy, even when applied under restricted and appropriate circumstances, emerges from a number of complex, interacting factors that are individually ignored, though each contributes to the overall measure (indicated by straight black arrows). This is not a problem when running economy is being used for the purpose of predicting running performance but it may become very important if we are using our model for a different purpose, for example to decide on the best training regimen to improve performance. At this point, the value of the model is much more influenced by the factors that were previously only an implicit part of the model. Two athletes may have limitations in their running economy for very different reasons, one due to biomechanical inefficiency arising from flaws in her posture or stride rate, another due to cardiovascular inefficiency. Each would be rectified by distinct training regimens (as indicated by blue, dotted arrows), which could not be chosen merely on the basis of running economy. It is the purpose of the model that specifies which components of reality need to be modelled and in how much detail.
Figure 1 The importance of specifying both the purpose of a model, and the mapping between the model components and aspects of the ‘real world’. Here, we illustrate the importance of two principles of modelling using the running analogy from Box 1 . It is tempting, when we have identified a reliable measure, one that seems to capture the essence of what we are trying to characterize and that has predictive value, to make it generally applicable. We may forget or ignore both the elements of the world that we are not including in our model and, furthermore, the ways in which the components that we are modelling may map to the real world. The figure illustrates how we might use ‘Running Economy’ (the volume of oxygen consumed at steady state running) as described in Box 1 as a predictor of race performance. Having the figures for oxygen consumption on two athletes allows a direct comparison of how they are likely to fare in competition. Indeed, even if the measurements were obtained at a range of speeds and on different terrain, the measures may be comparable. Running economy serves as a good model of running ability. But, as indicated in the figure, it only fulfils this purpose within predefined constraints. For example, it is standard to measure running economy at a pace that is below that at which a person is near their lactate threshold (the point at which oxygen consumption cannot keep up with demand) and therefore the model is inadequate if we wish to judge running ability for shorter distance sprints in which lactate levels are highly relevant ( Billat, 1996 ). Here the mapping between the model and reality has gone wrong because the model is applied for a setting where its assumptions are no longer relevant or valid. Unless we are explicit about our mapping between model components and reality and the working assumptions that justify it, our model will ultimately lead us astray. Such error lies not in the model but in how it is used. It should also be noted that running economy, even when applied under restricted and appropriate circumstances, emerges from a number of complex, interacting factors that are individually ignored, though each contributes to the overall measure (indicated by straight black arrows). This is not a problem when running economy is being used for the purpose of predicting running performance but it may become very important if we are using our model for a different purpose, for example to decide on the best training regimen to improve performance. At this point, the value of the model is much more influenced by the factors that were previously only an implicit part of the model. Two athletes may have limitations in their running economy for very different reasons, one due to biomechanical inefficiency arising from flaws in her posture or stride rate, another due to cardiovascular inefficiency. Each would be rectified by distinct training regimens (as indicated by blue, dotted arrows), which could not be chosen merely on the basis of running economy. It is the purpose of the model that specifies which components of reality need to be modelled and in how much detail.CC BY · PubMed Central

Verification

Claims the paper makes

0 Valid · 0 Not valid · 34 Not checked · 30 Own results and interpretations

27 more claims

Create a free account to see all 64 claims and to verify the 49 references still open.

See all claims — free account
  • Probably not a substantial claim. 26
  • It may state two separate claims as one. 15

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Citations

Cites

Show all 66 references

Research works printed but not linked: 2

  1. Dayan P Abbott LF . Theoretical neuroscience. Cambridge, MA: The MIT Press; 2001.
  2. Marr D . Vision. Cambridge, MA: The MIT Press; 1982.

Cited by

See all 64

Metadata

Authors
Teufel, C., & Fletcher, P. C.
Affiliations
Cardiff University (GB)
Corresponding author
Fletcher, Paul C.
Venue
Brain
Published
2016-08-20
Volume
139
Issue
10
Pages
2600-2608
ISSN
  • 0006-8950 (ISSN-L)
  • 1460-2156
OpenAlex
W2510894207
Cited by
64
References
66
Citations by year
2026: 6 · 2025: 5 · 2024: 5 · 2023: 6 · 2022: 2 · 2021: 5 · 2020: 4 · 2019: 5 · 2018: 19 · 2017: 7
FWCI
1.09
Type
journal article
Language
en
Open access
Licence
cc-by
Locations
  • Brain (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • ORCA Online Research @Cardiff (Cardiff University) (acceptedVersion)
  • HighWire Press Open Archive (submittedVersion)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Apollo (cc-by)
Funders
Wellcome Trust (093875/Z/10/Z, 093875)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Functional Brain Connectivity Studies
  • Social Sciences › Arts and Humanities › Philosophy › Mental Health and Psychiatry
Keywords
computational model, neurological disorders, mental illness, brain dysfunction, clinical neuroscience, model specification, diagnosis and treatment
MeSH terms
Mental Disorders, Brain Diseases, Computer Simulation, Humans, Nervous System Diseases, Computational Biology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-09-28, under version 3 of the claim definition (prompt extract-v3).

A second model labelled each claim on 2026-09-28 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: The promises and pitfalls of applying computational models to neurological and psychiatric disorders. https://app.cochat.ai/papers/10.1093/brain/aww209
The paper
Teufel, C., & Fletcher, P. C. (2016). The promises and pitfalls of applying computational models to neurological and psychiatric disorders. Brain. https://doi.org/10.1093/brain/aww209

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.