Neuro-computational mechanisms underlying the effects of anxiety and motivation on biased attentional and learning processes
Lead Research Organisation:
Goldsmiths University of London
Department Name: Psychology
Abstract
Computational models in combination with neuroimaging techniques are increasingly being applied to the study of psychiatric disorders. However, their application to healthy individuals that may represent a population at risk to develop these disorders remains unexplored. Here we propose to reveal the neuro-computational mechanisms subserving cognitive biases in healthy individuals that can generate serious and debilitating symptoms through anxiety and depressive disorders. Anxiety and depression are amongst the most common mental health problems worldwide, costing the UK an estimated £77 billion a year. By combining neuroscientific and computational modelling techniques, I aim to further our understanding of the mechanisms which construct cognitive biases within healthy individuals that lead to these problems in well-being.
An increasingly influential view proposes that the brain learns about its environment by comparing prior beliefs with sensory evidence, with the aim to generate increasingly accurate predictions about future states. This process can be best understood using computational models of Bayesian inference, which account for how the brain makes inferences about the state of its environment. Specifically, these models propose that by weighting both sensory data and prior belief according to their reliability (or precision), the brain can parsimoniously explain away incoming sensory information.
This proposal intends to show that the predictive nature of perception runs in tandem with our long-term memories, motivations and emotions, creating cognitive biases that can transform computational parsimony, into computational pathology.
Accordingly, I will investigate how everyday factors such as intrinsic motivation and anxiety can alter estimates of precision or reliability ascribed to prior belief or sensory data, thereby biasing perception and learning. Analysing neurophysiological and behavioural data in combination with Bayesian modelling will provide a mechanistic understanding of this previously unexplored non-optimal inference process. Moreover, by using a class of Bayesian models specifically introduced to assess individual differences, I ultimately aim to determine how inter-individual variation in estimates of precision can account for variability in individual learning under different degrees of anxiety and intrinsic motivation.
An increasingly influential view proposes that the brain learns about its environment by comparing prior beliefs with sensory evidence, with the aim to generate increasingly accurate predictions about future states. This process can be best understood using computational models of Bayesian inference, which account for how the brain makes inferences about the state of its environment. Specifically, these models propose that by weighting both sensory data and prior belief according to their reliability (or precision), the brain can parsimoniously explain away incoming sensory information.
This proposal intends to show that the predictive nature of perception runs in tandem with our long-term memories, motivations and emotions, creating cognitive biases that can transform computational parsimony, into computational pathology.
Accordingly, I will investigate how everyday factors such as intrinsic motivation and anxiety can alter estimates of precision or reliability ascribed to prior belief or sensory data, thereby biasing perception and learning. Analysing neurophysiological and behavioural data in combination with Bayesian modelling will provide a mechanistic understanding of this previously unexplored non-optimal inference process. Moreover, by using a class of Bayesian models specifically introduced to assess individual differences, I ultimately aim to determine how inter-individual variation in estimates of precision can account for variability in individual learning under different degrees of anxiety and intrinsic motivation.
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| ES/P00072X/1 | 01/10/2017 | 30/09/2027 | |||
| 1939637 | Studentship | ES/P00072X/1 | 01/10/2017 | 31/03/2021 | Thomas Hein |
| Description | Everyday temporary emotional experiences, like when feeling anxious about an upcoming event or performance, can create learning biases in healthy individuals. These can change how we learn from and interact with the world. We used a novel combination of computational models of learning and moment by moment analysis of responses from the brain. This allowed us to look in more detail at how changes to how we compute information while learning are associated with changes to neural processing. We found that states of anxiety can lead to reduced rates of learning and misestimations of uncertainty (uncertainty about if outcomes will be rewarding and uncertainty about if those rewards change over time). An example of which is an excessive sense of confidence in how well we can learn about rewards and their outcomes. Unfortunately, in our work, these excessive estimates of confidence under states of anxiety did not correspond to the true hidden nature of our learning task, and this resulted in poorer reward learning performance overall when compared to non-anxious individuals. We also discovered that when we feel anxious, we have difficulties learning from environments that change over time. Inducing state anxiety in healthy volunteers led to a misestimation of the uncertainty associated with environmental change, potentially linking research on the misinterpretation of uncertainty as distressing in those high in anxiety. These changes to how the brain computed uncertainty and reward were also associated with neural changes, with less representation of the signals corresponding to learning in our state anxious group than in the non-anxious group. A further discovery was that anxious states can impair learning in tasks requiring motor performance, like playing the piano. State anxiety in this reward-based motor learning task induced biases about the hidden performance goal and it's stability throughout time through changes to estimates of uncertainty and neural bursts of sensorimotor beta oscillations. These discoveries led to new research questions, such as: do misestimations of uncertainty under state anxiety, leading to temporary biases in beliefs, over time start to fit a profile of responses and biases found in clinically or highly anxious individuals? I believe the achievements of this ESRC funded research wholly met the grant objectives originally proposed. Our findings have already been useful to the broader scientific community researching computational psychiatry and how misestimations of uncertainty are linked to affective disorders. The latter linked to affective disorders and misestimated uncertainty led to collaborations with researchers at MIT and the Translational Neuromodeling Unit Institute for Biomedical Engineering University of Zurich opening up, and to invited speakers from UCL/California Institute of Technology presenting at our university at Goldsmiths London. I further hope that my research continues to be useful to the growing literature on computational learning mechanisms and to be helpful in understanding anxiety in both healthy and clinical populations. |
| Exploitation Route | This PhD was the first to provide a mechanistic understanding of how temporary anxious states impair reward-based learning in volatile environments. The results have implications for understanding cognitive biases and impaired learning in healthy individuals exposed to upcoming uncertain threats, but could also generalise to clinical settings. I envisage these results being used to inform research on how learning biases can start and persist at the computational level in the brain. I anticipate future work to link these findings to the development of computational biases. Further, I envisage future research will determine whether the distributed network of brain regions and neurotransmitter systems linked to anxiety-disorders and trait anxiety interact with the neural representation of hierarchical learning signals. Our data also imply that the combination of Bayesian learning models and analysis of brain oscillation rhythms can help better understand learning mechanisms in healthy individuals, the ways in which anxiety modulates learning, and how these relate to disorders of anxiety. A final broad envision of this work is that further pharmacological and therapeutical investigation can help to alleviate some of the distressing and uncomfortable experiences associated with anxiety. I imagine a data-driven future of computationally informed individually tailored diagnoses and treatment. |
| Sectors | Education,Healthcare,Pharmaceuticals and Medical Biotechnology |
| URL | https://www.sciencedirect.com/science/article/pii/S1053811920309095 |
| Description | Neural oscillations related to belief updating in state anxiety |
| Organisation | Massachusetts Institute of Technology |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | Designed the experiment, collected the data, analysed the data, wrote code for data analysis, wrote the manuscript, edited the manuscript. |
| Collaborator Contribution | Wrote code for data analysis, wrote the manuscript, edited the manuscript. |
| Impact | Publication in process. |
| Start Year | 2019 |
| Description | State Anxiety Biases Estimates of Uncertainty During Reward Learning in Volatile Environments |
| Organisation | ETH Zurich |
| Department | Institute for Biomedical Engineering |
| Country | Switzerland |
| Sector | Academic/University |
| PI Contribution | Designed the experiment, collected the data, analysed the data, wrote code for data analysis, wrote the manuscript, edited the manuscript. |
| Collaborator Contribution | Wrote code for data analysis. Help in improving writing manuscript. |
| Impact | One publication (in review process) |
| Start Year | 2018 |
| Description | Trait anxiety alters the neural oscillatory correlates of predictions and prediction errors during reward learning. |
| Organisation | National Research University Higher School of Economics |
| Country | Russian Federation |
| Sector | Academic/University |
| PI Contribution | We devised the experiment to answer questions from our eLife (2020) and Neuroimage (2021) and upcoming (under review) Journal of Neuroscience (2021) papers on sate anxiety. The purpose was to answer questions about the neural oscillations of the predictive coding framework in high trait anxious individuals using MEG, so better improve the fidelity of electrophysiological recordings in higher frequencies (a limitation of our EEG work). I wrote the code/programmed the experimental task. I later analysed all behavioural data, MEG electrophysiological data, and fit a computational model of learning to behavioural responses. I then wrote the manuscript for inclusion in my thesis and later publication. |
| Collaborator Contribution | Our contributors recorded the MEG data under the supervision of my primary supervisor Dr Maria Herrojo Ruiz. |
| Impact | Cognitive neuroscience. Computational neuroscience. |
| Start Year | 2019 |