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

The DOI matches its Crossref record; 50 of the 54 references it cites were found in a registry or an index; 30 of the 44 sources its claims cite were checked for support.

Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis

Open access

Abstract

Recent theories of cortical function construe the brain as performing hierarchical Bayesian inference. According to these theories, the precision of prediction errors plays a key role in learning and decision-making, is controlled by dopamine and contributes to the pathogenesis of psychosis. To test these hypotheses, we studied learning with variable outcome-precision in healthy individuals after dopaminergic modulation with a placebo, a dopamine receptor agonist bromocriptine or a dopamine receptor antagonist sulpiride (dopamine study n = 59) and in patients with early psychosis (psychosis study n = 74: 20 participants with first-episode psychosis, 30 healthy controls and 24 participants with at-risk mental state attenuated psychotic symptoms). Behavioural computational modelling indicated that precision weighting of prediction errors benefits learning in health and is impaired in psychosis. FMRI revealed coding of unsigned prediction errors, which signal surprise, relative to their precision in superior frontal cortex (replicated across studies, combined n = 133), which was perturbed by dopaminergic modulation, impaired in psychosis and associated with task performance and schizotypy (schizotypy correlation in 86 healthy volunteers). In contrast to our previous work, we did not observe significant precision-weighting of signed prediction errors, which signal valence, in the midbrain and ventral striatum in the healthy controls (or patients) in the psychosis study. We conclude that healthy people, but not patients with first-episode psychosis, take into account the precision of the environment when updating beliefs. Precision weighting of cortical prediction error signals is a key mechanism through which dopamine modulates inference and contributes to the pathogenesis of psychosis.

Overview

CoChat Summary

TL;DR

Precision weighting of cortical unsigned prediction error signals in superior frontal cortex/dorsal anterior cingulate benefits learning, is modulated by dopamine, and is impaired in first-episode psychosis. The authors conclude this mechanism may contribute to psychotic symptom pathogenesis, while signed prediction error precision weighting was not replicated in the psychosis study.

Key findings

  • Unsigned prediction errors were coded in bilateral superior frontal cortex (SFC) and dorsal anterior cingulate cortex/pre-SMA, and were precision-weighted; in the dopamine study, sulpiride significantly perturbed precision weighting in left SFC (interaction F{2,56} = 4.025, p = 0.023), with less precision weighting than placebo (F{2,37} = 5.44, p = 0.025).
  • Greater cortical precision weighting correlated with better task performance (left SFC Rho = −0.45, p < 0.001; right SFC Rho = −0.40, p = 0.002; dACC Rho = −0.25, p = 0.055).
  • In the psychosis study, first-episode psychosis (FEP) showed decreased performance and less benefit from more precise information; FEP was best fit by a simple Rescorla–Wagner model without precision weighting, unlike healthy controls and at-risk mental state participants.
  • FEP had diminished precision weighting of unsigned prediction error in right SFC compared with controls (T = 3.38, voxels = 9, p = 0.035), and higher schizotypy related to less right SFC precision weighting in healthy volunteers (Rho = −0.25, p = 0.024). No significant signed prediction error coding, precision weighting, or precision-by-group interaction was found in ventral striatum or midbrain (all p > 0.1).

Methods

  • Dopamine study: 59 healthy volunteers received double-blind single-dose placebo, D2-antagonist sulpiride (600 mg), or dopamine agonist bromocriptine (2.5 mg) before fMRI.
  • Psychosis study: 74 participants total — 20 first-episode psychosis, 30 healthy controls, and 24 at-risk mental state — completed fMRI and behavioural testing.
  • Task: participants learned reward distributions differing in precision (dopamine study SDs 5, 10, 15; psychosis study SDs 5 or 15); analyses used behavioural computational modelling (Rescorla–Wagner and Pearce–Hall models) and fMRI parametric modulation of unsigned prediction errors, with ROI analyses in SFC/dACC.

Limitations

  • The authors state they did not replicate previous findings of significant precision-weighting of signed prediction errors in midbrain and ventral striatum in controls or patients, so they could not test whether signed prediction error precision differs in patients versus controls.
  • The authors note the psychosis study used a different MRI scanner, head coil, and imaging sequence than the dopamine study, and had slight design differences, which may have reduced sensitivity to subcortical/midbrain signed prediction error precision weighting.
  • The authors acknowledge modest patient sample size and marginal significance of the symptom correlation, and state that dopamine function was not measured in the clinical studies, leaving possible non-dopaminergic mechanisms for patient deficits.

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

Integrity signals

Registry

Crossref · 10.1038/s41380-020-0803-8

Cites

50 of 50 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Example of a trial. The participants were instructed to learn the mean of a reward distribution. First a fixation cross was presented after which the participants were informed about the standard deviation (which indicated the precision) of the reward distribution. Subsequently the participants were asked to make a prediction regarding the upcoming reward, which was presented to the participant in combination with the prediction error (in yellow) after an anticipation period.
Fig. 1 Example of a trial. The participants were instructed to learn the mean of a reward distribution. First a fixation cross was presented after which the participants were informed about the standard deviation (which indicated the precision) of the reward distribution. Subsequently the participants were asked to make a prediction regarding the upcoming reward, which was presented to the participant in combination with the prediction error (in yellow) after an anticipation period.CC BY · PubMed Central
Fig. 2 Behavioural results for the dopamine study. a–e display the average learning curves, reflecting the absolute distance between the actual mean of the distribution and participants’ estimate of the mean distribution over 30 trials averaged over the three sessions for each participant. The distance between the prediction and the actual mean of the distribution defines performance error. Therefore, lower values of performance error are better. Asterisks indicate Bonferroni-corrected significant differences across conditions. a Performance was significantly better in the high-precision condition compared with the low-precision condition when combing all participants, especially at the beginning of the experiment. b There were no clear differences between groups when analysing trial-by-trial performance. The placebo (c), sulpiride (d) and bromocriptine (e) group showed only a significant difference between the precision conditions one or two trials in the beginning of the experiment. f Averaging performance error across all trials, we see overall better performance in the placebo condition compared with the sulpiride condition, and higher performance in more precision conditions. g Initial learning rate parameters are displayed here for the Pearce–Hall model. Initial learning rates were higher in the Bromocriptine condition compared with the sulpiride condition. h Learning rate decay parameters from the Pearce–Hall model are displayed here. Precision and group did not affect the learning rate decay parameter. i No differences were found in number of missed trials, and k scrolling distance. j Reaction times were quicker in the sulpiride group. Error bars represent standard error of the mean.
Fig. 2 Behavioural results for the dopamine study. a–e display the average learning curves, reflecting the absolute distance between the actual mean of the distribution and participants’ estimate of the mean distribution over 30 trials averaged over the three sessions for each participant. The distance between the prediction and the actual mean of the distribution defines performance error. Therefore, lower values of performance error are better. Asterisks indicate Bonferroni-corrected significant differences across conditions. a Performance was significantly better in the high-precision condition compared with the low-precision condition when combing all participants, especially at the beginning of the experiment. b There were no clear differences between groups when analysing trial-by-trial performance. The placebo (c), sulpiride (d) and bromocriptine (e) group showed only a significant difference between the precision conditions one or two trials in the beginning of the experiment. f Averaging performance error across all trials, we see overall better performance in the placebo condition compared with the sulpiride condition, and higher performance in more precision conditions. g Initial learning rate parameters are displayed here for the Pearce–Hall model. Initial learning rates were higher in the Bromocriptine condition compared with the sulpiride condition. h Learning rate decay parameters from the Pearce–Hall model are displayed here. Precision and group did not affect the learning rate decay parameter. i No differences were found in number of missed trials, and k scrolling distance. j Reaction times were quicker in the sulpiride group. Error bars represent standard error of the mean.CC BY · PubMed Central
Fig. 3 Brain imaging results for the dopamine modulation study. a Unsigned prediction errors were coded in bilateral superior frontal cortex and dorsal anterior cingulate cortex. The left side of the brain is the left side of the image. b–d When exploring these regions further, we find that unsigned prediction errors are coded in a precision-weighted fashion as indicated by the strong unsigned prediction error signal in the high-precision condition, which declines over the medium- and low-precision condition in the placebo and bromocriptine group. Importantly, sulpiride perturbed precision-weighting significantly in the left SFC. Error bars represent standard error of the mean.
Fig. 3 Brain imaging results for the dopamine modulation study. a Unsigned prediction errors were coded in bilateral superior frontal cortex and dorsal anterior cingulate cortex. The left side of the brain is the left side of the image. b–d When exploring these regions further, we find that unsigned prediction errors are coded in a precision-weighted fashion as indicated by the strong unsigned prediction error signal in the high-precision condition, which declines over the medium- and low-precision condition in the placebo and bromocriptine group. Importantly, sulpiride perturbed precision-weighting significantly in the left SFC. Error bars represent standard error of the mean.CC BY · PubMed Central
Fig. 4 Cortical learning signals and performance. Precision weighting of unsigned prediction errors in the left SFC (a) and right SFC (b) correlates with performance (i.e., difference between mean of the reward distribution and predicted mean) on the task.
Fig. 4 Cortical learning signals and performance. Precision weighting of unsigned prediction errors in the left SFC (a) and right SFC (b) correlates with performance (i.e., difference between mean of the reward distribution and predicted mean) on the task.CC BY · PubMed Central
Fig. 5 Behavioural results for the patient study; green bars are healthy controls, blue ARMS and orange first-episode psychosis. a–e display the average learning curves, reflecting the absolute distance between the actual mean of the distribution and participants’ estimate of the mean distribution over 30 trials averaged over the three sessions for each participant. The distance between the prediction and the actual mean of the distribution defines performance error. Therefore, lower values of performance are better. Asterisks indicate Bonferroni-corrected significant differences across conditions. a Performance was significantly better in the high-precision condition compared with the low-precision condition when combing all participants, especially at the beginning of the experiment. b HCS performed better than FEP, but not compared with ARMS. The colour of the asterisk indicates a significant difference of the patient group with HCS. Healthy controls (c) showed a significant difference between the precision conditions, whereas the patient groups did not (d ARMS, e FEP). f Averaging performance error across all trials, HCS and ARMS perform better than FEP, and benefitted from more precise information, whereas FEP did not. g Initial learning rates are displayed here for the Pearce–Hall model. Learning rates were higher in the high-precision condition compared with the low-precision condition in HCS and ARMS, but not for FEPS, although an interaction was not significant. h Learning rate decay parameters are displayed here for the Pearce–Hall model. Precision did not affect the learning rate decay parameter. i HCS had slightly fewer missed trials than FEP. j Reaction times were equal across groups. k Scrolling distance was equal across groups. Error bars represent standard error of the mean.
Fig. 5 Behavioural results for the patient study; green bars are healthy controls, blue ARMS and orange first-episode psychosis. a–e display the average learning curves, reflecting the absolute distance between the actual mean of the distribution and participants’ estimate of the mean distribution over 30 trials averaged over the three sessions for each participant. The distance between the prediction and the actual mean of the distribution defines performance error. Therefore, lower values of performance are better. Asterisks indicate Bonferroni-corrected significant differences across conditions. a Performance was significantly better in the high-precision condition compared with the low-precision condition when combing all participants, especially at the beginning of the experiment. b HCS performed better than FEP, but not compared with ARMS. The colour of the asterisk indicates a significant difference of the patient group with HCS. Healthy controls (c) showed a significant difference between the precision conditions, whereas the patient groups did not (d ARMS, e FEP). f Averaging performance error across all trials, HCS and ARMS perform better than FEP, and benefitted from more precise information, whereas FEP did not. g Initial learning rates are displayed here for the Pearce–Hall model. Learning rates were higher in the high-precision condition compared with the low-precision condition in HCS and ARMS, but not for FEPS, although an interaction was not significant. h Learning rate decay parameters are displayed here for the Pearce–Hall model. Precision did not affect the learning rate decay parameter. i HCS had slightly fewer missed trials than FEP. j Reaction times were equal across groups. k Scrolling distance was equal across groups. Error bars represent standard error of the mean.CC BY · PubMed Central
Fig. 6 Brain imaging results for the psychosis study. a Precision weighting of prediction error in superior frontal cortex region of interest. Precision weighting is significantly diminished in first-episode psychosis. The y-axis provides beta estimates for the unsigned prediction error in different precision conditions in arbitrary units (a.u.). b Diminished precision-weighting of prediction error is correlated with positive symptoms. Error bars represent standard error of the mean.
Fig. 6 Brain imaging results for the psychosis study. a Precision weighting of prediction error in superior frontal cortex region of interest. Precision weighting is significantly diminished in first-episode psychosis. The y-axis provides beta estimates for the unsigned prediction error in different precision conditions in arbitrary units (a.u.). b Diminished precision-weighting of prediction error is correlated with positive symptoms. Error bars represent standard error of the mean.CC BY · PubMed Central
Fig. 7 Cortical learning signals and schizotypy. Higher SPQ scores (schizotypy) were correlated with less precision-weighting of the unsigned prediction error signal in the right superior frontal cortex, including when controlling for experimental group; Rho = −0.25, p = 0.024.
Fig. 7 Cortical learning signals and schizotypy. Higher SPQ scores (schizotypy) were correlated with less precision-weighting of the unsigned prediction error signal in the right superior frontal cortex, including when controlling for experimental group; Rho = −0.25, p = 0.024.CC BY · PubMed Central

Verification

Claims the paper makes

11 Valid · 2 Not valid · 4 Not checked · 93 Own results and interpretations

70 more claims

Create a free account to see all 110 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 22
  • It may state two separate claims as one. 18

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.

5 of 6 · the rest aren’t loaded

1 more finding

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Show all 54 references

Cited by

See all 113

Metadata

Authors
Haarsma, J., Fletcher, P. C., Griffin, J. D., Taverne, H. J., Ziauddeen, H., Spencer, T. J., Miller, C., Katthagen, T., Goodyer, I., Diederen, K. M. J., & Murray, G. K.
Affiliations
  • University of Cambridge (GB)
  • King's College London (GB)
  • Humboldt-Universität zu Berlin (DE)
Corresponding author
  • Diederen, K. M. J.
  • Murray, G. K.
Venue
Molecular Psychiatry
Published
2020-06-24
Volume
26
Issue
9
Pages
5320-5333
ISSN
  • 1359-4184 (ISSN-L)
  • 1476-5578
OpenAlex
W3036131087
Cited by
108
References
54
Citations by year
2026: 20 · 2025: 12 · 2024: 19 · 2023: 21 · 2022: 17 · 2021: 12 · 2020: 7
FWCI
10.38
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Psychiatry (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Research Portal (King's College London) (submittedVersion, cc-by)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Apollo
Funders
  • Wellcome Trust (095844/Z/11/Z, 095844, 093270)
  • National Institute for Health and Care Research
  • University of Cambridge
  • University College London (095844/Z/11/Z)
  • NIHR Cambridge Biomedical Research Centre
Topics
  • Health Sciences › Medicine › Psychiatry and Mental health › Schizophrenia research and treatment
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
  • Life Sciences › Neuroscience › Cellular and Molecular Neuroscience › Neurotransmitter Receptor Influence on Behavior
Keywords
dopamine, psychosis, first-episode psychosis, Bayesian inference, learning, surprise, superior frontal cortex, schizotypy, dopaminergic modulation, bromocriptine, sulpiride, clinical high risk for psychosis, computational modelling
MeSH terms
Bayes Theorem, Brain, Dopamine, Humans, Learning, Magnetic Resonance Imaging, Psychotic Disorders, Reward

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 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (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: Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis. https://app.cochat.ai/papers/10.1038/s41380-020-0803-8
The paper
Haarsma, J., Fletcher, P. C., Griffin, J. D., Taverne, H. J., Ziauddeen, H., Spencer, T. J., Miller, C., Katthagen, T., Goodyer, I., Diederen, K. M. J., & Murray, G. K. (2021). Precision weighting of cortical unsigned prediction error signals benefits learning, is mediated by dopamine, and is impaired in psychosis. Molecular Psychiatry. https://doi.org/10.1038/s41380-020-0803-8

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.