Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 77 of the 77 references it cites were found in a registry or an index; 76 of the 104 sources its claims cite were checked for support.

Predictive coding and neurocomputational psychiatry: a mechanistic framework for understanding mental disorders

Open access
Shaw, Alexander D., Sumner, Rachael L., Berndt, Lioba C. S. · 2026 · Frontiers in Psychiatry

Abstract

Predictive coding offers a powerful computational framework for understanding brain function and psychiatric disorders at a mechanistic level. This perspective synthesizes advances in computational psychiatry, proposing that mental disorders can be conceptualized as specific alterations in the brain's predictive inference machinery. We first outline the theoretical foundations of predictive coding, including Bayesian inference, free-energy minimization, and neural population dynamics, illustrating how these abstract computational principles map onto specific neural circuits and biophysical mechanisms. We then argue that diverse psychiatric conditions can be understood within this unified framework. Taken together, these links between theory, generative models and empirical data suggest a route by which predictive coding might be rendered a testable, modifiable, falsifiable construct within biological psychiatry. Beyond offering conceptual clarity, this framework has significant clinical implications, including the development of mechanistic biomarkers, personalized treatment approaches based on computational phenotypes, and novel therapeutic interventions targeting specific inferential abnormalities. By grounding psychiatric symptoms in aberrant predictive processes implemented in neural circuitry, this approach promises a more mechanistic understanding of mental disorders and a path toward more targeted, effective interventions.

Overview

CoChat Summary

## TL;DR

This review argues that predictive coding provides a mechanistic framework for psychiatry by linking symptoms to abnormalities in Bayesian inference, precision-weighting, and circuit-level prediction error processing. It presents schizophrenia, autism, depression, OCD, anxiety, and bipolar disorder as different patterns of inferential dysfunction and suggests implications for biomarkers, personalized treatment, and novel interventions.

## Key findings

  • Predictive coding is framed as a brain-inference model in which top-down predictions and bottom-up prediction errors are updated to minimize surprise/free energy; the paper links this to neural circuitry, mean-field models, and biophysical models such as Hodgkin–Huxley.
  • The paper proposes disorder-specific computational interpretations:
  • schizophrenia: aberrant precision-weighting of priors and prediction errors, with evidence including reduced mismatch negativity and diminished feedback connectivity in DCM studies;
  • autism: weak priors / low-precision priors leading to sensory overweighting and hypersensitivity;
  • depression: maladaptive attractor states and failure to update from positive evidence;
  • OCD: overactive error signals and excessive precision on prediction errors;
  • anxiety disorders: exaggerated uncertainty and threat predictions;
  • bipolar disorder: instability of precision and network dynamics across manic and depressive states.
  • For schizophrenia, the paper reports a DCM study in which patients had “markedly reduced backward connectivity from frontal cortex to visual cortex” and failed to increase this top-down connectivity when stimuli became predictable.
  • The paper argues that predictive coding may support mechanistic biomarkers, personalized treatment selection, and interventions such as DBS, closed-loop stimulation, biofeedback, exposure therapy, cognitive restructuring, and pharmacological re-tuning of neuromodulatory gain.

## Methods

  • Not an empirical study; this is a theoretical/review paper synthesizing predictive coding, Bayesian inference, free-energy minimization, and computational psychiatry.
  • The paper draws on prior empirical literature, including M/EEG, fMRI, DCM, and neurochemical findings, but it does not report a new sample or new data collection.
  • The main analysis is conceptual and integrative: it maps predictive-coding quantities (predictions, prediction errors, precision) onto psychiatric symptoms and cited circuit-level evidence.

## Limitations

  • The paper states that several limitations remain: integrating computational insights into clinical taxonomies is difficult, comorbidity complicates diagnosis, and current biomarkers grounded in predictive coding (e.g. mismatch negativity) have not yet shown sufficient specificity or reliability for clinical translation.
  • It also states that many predictive-coding quantities overlap with other frameworks such as reinforcement learning, Bayesian decision theory, and control-theoretic models, limiting the discriminative power of predictive coding unless it makes distinct empirical predictions.
  • Evident from the text: the paper is largely a conceptual synthesis rather than a direct test of its claims, so many disorder mappings are presented as hypotheses or interpretations rather than newly demonstrated findings.

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

Integrity signals

Registry

Crossref · 10.3389/fpsyt.2025.1713833

Cites

77 of 77 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Computational models, here exemplified by predictive coding, as a bridge between psychiatric symptoms and neurobiological mechanisms. Traditional approaches (grey dashed line) attempt to map clinical phenomena directly onto biological substrates, often yielding limited mechanistic insight. Predictive coding provides a principled intermediate layer (orange arrows), allowing symptoms to be reframed as computational inference failures-such as aberrant precision-weighting or dysfunctional attractor dynamics-which can in turn be grounded in specific circuit-level or receptor-level pathophysiology. Other computational frameworks (e.g. reinforcement-learning or control-theoretic models) can also occupy this intermediate level. Examples illustrate this tri-level mapping for hallucinations, sensory overload, anhedonia, and compulsions.
Figure 1 Computational models, here exemplified by predictive coding, as a bridge between psychiatric symptoms and neurobiological mechanisms. Traditional approaches (grey dashed line) attempt to map clinical phenomena directly onto biological substrates, often yielding limited mechanistic insight. Predictive coding provides a principled intermediate layer (orange arrows), allowing symptoms to be reframed as computational inference failures-such as aberrant precision-weighting or dysfunctional attractor dynamics-which can in turn be grounded in specific circuit-level or receptor-level pathophysiology. Other computational frameworks (e.g. reinforcement-learning or control-theoretic models) can also occupy this intermediate level. Examples illustrate this tri-level mapping for hallucinations, sensory overload, anhedonia, and compulsions.CC BY · PubMed Central
Figure 2 Schematic of hierarchical predictive coding across cortical levels. Left panel: Higher areas (N+1) send predicted signals (feedback, blue arrows) to lower sensory areas (N), forming empirical priors for incoming data. At the lower level, the difference between actual input (green arrow) and the top-down prediction constitutes a prediction error. This error is weighted (purple arrow) and passed forward (feedforward) to update the higher-level representation (red arrow) in the form of a ‘precision weighted prediction error’. Right panel: The encoding of predictions and prediction errors plausibly takes place in deep (L5/6) and superficial (L2/3) pyramidal populations, respectively. Furthermore, since this process is dynamical (rather than fixed) these computations can be formulated as differential equations (far right).
Figure 2 Schematic of hierarchical predictive coding across cortical levels. Left panel: Higher areas (N+1) send predicted signals (feedback, blue arrows) to lower sensory areas (N), forming empirical priors for incoming data. At the lower level, the difference between actual input (green arrow) and the top-down prediction constitutes a prediction error. This error is weighted (purple arrow) and passed forward (feedforward) to update the higher-level representation (red arrow) in the form of a ‘precision weighted prediction error’. Right panel: The encoding of predictions and prediction errors plausibly takes place in deep (L5/6) and superficial (L2/3) pyramidal populations, respectively. Furthermore, since this process is dynamical (rather than fixed) these computations can be formulated as differential equations (far right).CC BY · PubMed Central
Figure 3 Conceptual illustration of differences in neural dynamics between healthy (leftpanel) and depressed (right panel) states using a generic attractor framework. The system shown is a toy dynamical model for intuition, not a realistic simulation of a particular brain circuit. The three axes represent key computational variables in predictive coding: X-axis: Precision-weighted prediction error (how much sensory input updates beliefs). Y-axis: Top-down priors (strength of internal expectations, including cognitive biases). Z-axis: Neuromodulatory gain control (influence of neurotransmitters like NMDA on precision). In the healthy state (blue attractor), neural dynamics are chaotic and flexible, allowing for continuous updating of beliefs in response to sensory input. The system explores a broad range of states, reflecting an adaptive balance between prior beliefs and prediction errors. In contrast, the depressed state (red attractor) exhibits a collapsed, deep attractor basin, indicative of rigid, maladaptive inference. Here, overweighted priors (high Y) dominate, while prediction error updates are suppressed (low X), leading to inflexible belief updating. Additionally, reduced neuromodulatory gain (low Z) diminishes the system’s ability to dynamically adjust sensory precision, mirroring reduced serotonergic and dopaminergic function in depression.
Figure 3 Conceptual illustration of differences in neural dynamics between healthy (leftpanel) and depressed (right panel) states using a generic attractor framework. The system shown is a toy dynamical model for intuition, not a realistic simulation of a particular brain circuit. The three axes represent key computational variables in predictive coding: X-axis: Precision-weighted prediction error (how much sensory input updates beliefs). Y-axis: Top-down priors (strength of internal expectations, including cognitive biases). Z-axis: Neuromodulatory gain control (influence of neurotransmitters like NMDA on precision). In the healthy state (blue attractor), neural dynamics are chaotic and flexible, allowing for continuous updating of beliefs in response to sensory input. The system explores a broad range of states, reflecting an adaptive balance between prior beliefs and prediction errors. In contrast, the depressed state (red attractor) exhibits a collapsed, deep attractor basin, indicative of rigid, maladaptive inference. Here, overweighted priors (high Y) dominate, while prediction error updates are suppressed (low X), leading to inflexible belief updating. Additionally, reduced neuromodulatory gain (low Z) diminishes the system’s ability to dynamically adjust sensory precision, mirroring reduced serotonergic and dopaminergic function in depression.CC BY · PubMed Central

Verification

Claims the paper makes

54 Valid · 17 Not valid · 22 Not checked · 116 Own results and interpretations

105 more claims · 40 with their cited sources checked

Create a free account to see all 209 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 58
  • Probably not a substantial claim. 49

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 77 references

Research works printed but not linked: 2

  1. Beck, 1979, Cognitive therapy and the emotional disorders
  2. 2022, Diagnostic and statistical manual of mental disorders

Cited by

See all 9

Metadata

Authors
Shaw, A. D., Sumner, R. L., & Berndt, L. C. S.
Affiliations
  • University of Exeter (GB)
  • Auckland University of Technology (NZ)
Corresponding author
Berndt, Lioba C. S.
Venue
Frontiers in Psychiatry
Published
2026-01-07
Volume
16
Pages
1713833-1713833
ISSN
1664-0640 (ISSN-L)
OpenAlex
W7118698301
Cited by
9
References
77
Citations by year
2026: 8
FWCI
33.05
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Psychiatry (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • Explore Bristol Research (publishedVersion, cc-by)
  • Tuwhera (Auckland University of Technology) (submittedVersion, cc-by)
  • Bristol Research (University of Bristol) (submittedVersion, cc-by)
Funders
Wellcome
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
  • Social Sciences › Arts and Humanities › Philosophy › Mental Health and Psychiatry
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
Keywords
predictive coding, computational psychiatry, Bayesian inference, free energy principle, neural population dynamics, mental illness, mechanistic biomarkers, computational phenotypes, personalized treatment, therapeutic interventions, neural circuits

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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: Predictive coding and neurocomputational psychiatry: a mechanistic framework for understanding mental disorders. https://app.cochat.ai/papers/10.3389/fpsyt.2025.1713833
The paper
Shaw, A. D., Sumner, R. L., & Berndt, L. C. S. (2026). Predictive coding and neurocomputational psychiatry: a mechanistic framework for understanding mental disorders. Frontiers in Psychiatry. https://doi.org/10.3389/fpsyt.2025.1713833

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.