MRC Transition Support. CDA. Jill O'Reilly.
Lead Research Organisation:
UNIVERSITY OF OXFORD
Department Name: Experimental Psychology
Abstract
My research is concerned with how our brains build models of the environment in which we live, and use those models to make decisions and plan behaviour.
My research group's approach to understanding the brain is to use computational modelling. We develop experimental games that simulate problems people face in the real world, such as choosing between two options that cannot be clearly seen. Then we code algorithms (implemented in computer programmes) that solve the tasks in different ways, and test which algorithm is the best fit to human behaviour. Next, we use our knowledge of brain circuits to work out how these algorithms could be computed by populations of neurons in a real brain. I measure brain activity in humans using non-invasive imaging techniques such as MRI.
Some of my models also make predictions about how different brain chemicals (neurotransmitters and neuromodulators) affect the processing of information by the brain. To test these hypotheses, I will give healthy volunteers small doses of drugs that affect neurotransmitter concentrations and observe the resulting changes in their behaviour and brain activity.
A major technical challenge for my research programme is that we cannot simply record the activity of neurons in the human brain, without opening up the skull. To test our hypotheses in non-invasive experiments, we need work out how the modelled activity of neurons will be reflected in signals we can measure using non-invasive techniques like brain scanning (functional magnetic resonance imaging or magnetoencephalography). During the last four years, we have spent most of our time on solving this problem and are now confident that we have good models. However, we have not yet been able to complete the planned drug studies, which require a lot of slow and careful work to be done safely.
I am now applying for additional funding for my research programme to help me complete the proposed research. The bulk of the funding is to retain a key staff member in my team. Some of the funding is to buy out a percentage of my time from teaching so I am able to spend more time focussing on research and supporting my team.
My research group's approach to understanding the brain is to use computational modelling. We develop experimental games that simulate problems people face in the real world, such as choosing between two options that cannot be clearly seen. Then we code algorithms (implemented in computer programmes) that solve the tasks in different ways, and test which algorithm is the best fit to human behaviour. Next, we use our knowledge of brain circuits to work out how these algorithms could be computed by populations of neurons in a real brain. I measure brain activity in humans using non-invasive imaging techniques such as MRI.
Some of my models also make predictions about how different brain chemicals (neurotransmitters and neuromodulators) affect the processing of information by the brain. To test these hypotheses, I will give healthy volunteers small doses of drugs that affect neurotransmitter concentrations and observe the resulting changes in their behaviour and brain activity.
A major technical challenge for my research programme is that we cannot simply record the activity of neurons in the human brain, without opening up the skull. To test our hypotheses in non-invasive experiments, we need work out how the modelled activity of neurons will be reflected in signals we can measure using non-invasive techniques like brain scanning (functional magnetic resonance imaging or magnetoencephalography). During the last four years, we have spent most of our time on solving this problem and are now confident that we have good models. However, we have not yet been able to complete the planned drug studies, which require a lot of slow and careful work to be done safely.
I am now applying for additional funding for my research programme to help me complete the proposed research. The bulk of the funding is to retain a key staff member in my team. Some of the funding is to buy out a percentage of my time from teaching so I am able to spend more time focussing on research and supporting my team.
Technical Summary
The research in the CDA aims to develop a computational characterization of the roles of different brain systems in the formation and tuning of predictive models of the environment, which preidcts signals measurable with human neuroimaging.
The challenges involved in the research were: [Step 1] to design suitable tasks and models for which the necessary computations could be clearly defined at an algorithmic level; [Step 2] to posit circuit-level mechanisms that could underlie the computations; [Step 3] to build physiologically specified (biophysical) models of those mechanisms; [Step 4] to conduct drug studies that are safe, ethical and sufficiently powered to test the hypotheses.
So far, we have made good progress with steps 1-3, establishing a variety of paradigms and models in which predictions at the cells-and-circuits level can really be tested in human neuroimaging, and we have found some interesting results. However, we have only recently started to attempt drug manipulations (mid-data collection for one study; two more studies planned).
Although in some respects I have made good progress (one joint first-author paper in Science and one last author paper in eLife), I do not quite have enough last-author papers to secure a major research grant to follow on from the CDA. However, a further three major last-author studies are in the data analysis/write-up stage. Once these studies are published, I would hope to be in a much better position to apply for further funding.
My lab has experienced significant disruption over the CDA period including three moves of location (two involuntary) and two periods of maternity leave (for me).
The crucial benefits of Transition Funding would be to protect 50% of my time for research (I have a permanent contract but it will involve a lot of teaching), so I can be present in the lab and focus on completing projects and writing grants, and to allow me to retain my post doc which is essential for continuity.
The challenges involved in the research were: [Step 1] to design suitable tasks and models for which the necessary computations could be clearly defined at an algorithmic level; [Step 2] to posit circuit-level mechanisms that could underlie the computations; [Step 3] to build physiologically specified (biophysical) models of those mechanisms; [Step 4] to conduct drug studies that are safe, ethical and sufficiently powered to test the hypotheses.
So far, we have made good progress with steps 1-3, establishing a variety of paradigms and models in which predictions at the cells-and-circuits level can really be tested in human neuroimaging, and we have found some interesting results. However, we have only recently started to attempt drug manipulations (mid-data collection for one study; two more studies planned).
Although in some respects I have made good progress (one joint first-author paper in Science and one last author paper in eLife), I do not quite have enough last-author papers to secure a major research grant to follow on from the CDA. However, a further three major last-author studies are in the data analysis/write-up stage. Once these studies are published, I would hope to be in a much better position to apply for further funding.
My lab has experienced significant disruption over the CDA period including three moves of location (two involuntary) and two periods of maternity leave (for me).
The crucial benefits of Transition Funding would be to protect 50% of my time for research (I have a permanent contract but it will involve a lot of teaching), so I can be present in the lab and focus on completing projects and writing grants, and to allow me to retain my post doc which is essential for continuity.
Planned Impact
I have nothing specific to Transition Support to add here, except to say that the Transition Support would allow me to realise the full impact of the research by completing and publishing work in a timely manner.
Organisations
Publications
Barron HC
(2020)
Neuronal Computation Underlying Inferential Reasoning in Humans and Mice.
in Cell
Holton E
(2024)
Goal commitment is supported by vmPFC through selective attention.
in Nature human behaviour
Kaanders P
(2021)
Medial Frontal Cortex Activity Predicts Information Sampling in Economic Choice.
in The Journal of neuroscience : the official journal of the Society for Neuroscience
Kaanders P
(2021)
Dissociable mechanisms of information sampling in prefrontal cortex and the dopaminergic system
in Current Opinion in Behavioral Sciences
Klaassen FH
(2021)
Defensive freezing and its relation to approach-avoidance decision-making under threat.
in Scientific reports
Koolschijn R
(2024)
Noradrenaline causes a spread of association in the hippocampal cognitive map
Lawrance EL
(2022)
The Computational and Neural Substrates of Ambiguity Avoidance in Anxiety.
in Computational psychiatry (Cambridge, Mass.)
Marshall T
(2021)
The representation of priors and decisions in parietal cortex
Marshall TR
(2024)
The representation of priors and decisions in the human parietal cortex.
in PLoS biology
| Description | BBSRC SLOLA |
| Amount | £2,911,000 (GBP) |
| Funding ID | BB/W003392/1 |
| Organisation | Biotechnology and Biological Sciences Research Council (BBSRC) |
| Sector | Public |
| Country | United Kingdom |
| Start | 03/2022 |
| End | 04/2027 |
| Description | Inhibitory engrams in learning and memory consolidation (research grant) |
| Amount | £613,000 (GBP) |
| Funding ID | MR/W01971X/1 |
| Organisation | Medical Research Council (MRC) |
| Sector | Public |
| Country | United Kingdom |
| Start | 03/2022 |
| End | 04/2025 |
| Title | Data from: Control of entropy in neural models of environmental state |
| Description | Humans and animals construct internal models of their environment in order to select appropriate courses of action. The representation of uncertainty about the current state of the environment is a key feature of these models that controls the rate of learning as well as directly affecting choice behaviour. To maintain flexibility, given that uncertainty naturally decreases over time, most theoretical inference models include a dedicated mechanism to drive up model uncertainty. Here we probe the long-standing hypothesis that noradrenaline is involved in determining the entropy, and thus flexibility, of neural models. Pupil diameter, which indexes neuromodulatory state including noradrenaline release, predicted increases (but not decreases) in entropy in a neural state model encoded in human medial orbitofrontal cortex, as measured using multivariate functional MRI. Activity in anterior cingulate cortex predicted pupil diameter. These results provide evidence for top-down, neuromodulatory control of entropy in neural state models. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2019 |
| Provided To Others? | Yes |
| URL | https://datadryad.org/stash/dataset/doi:10.5061/dryad.jk17vk0 |
| Title | Decision-making in dynamic, continuously evolving environments: Quantifying the flexibility of human choice |
| Description | # Decision-making in dynamic, continuously evolving environments: Quantifying the flexibility of human choice [https://doi.org/10.5061/dryad.02v6wwq6b](https://doi.org/10.5061/dryad.02v6wwq6b) This repository contains EEG and behavioural data from a study of continuous decision making in a random-dot kinteogram paradigm, in which participants aim to detect consistent periods of motion ('response periods') in background noise. Participants complete 6 blocks, each consisting of 4 conditions (different 'environments'). Each condition lasts 5 minutes. Full details of the paradigm are given in the preprint at [https://www.biorxiv.org/content/10.1101/2022.08.18.504278.abstract](https://www.biorxiv.org/content/10.1101/2022.08.18.504278.abstract). *** ## Description of the data and file structure The location of data files in the repository are highlighted in **bold**, and the relevant MATLAB functions to load/explore the data are given in *italics*. Code to analyse this data is stored at our online code respository, [https://github.com/CCNHuntLab/ruesseler-eeg-analysis](https://github.com/CCNHuntLab/ruesseler-eeg-analysis). ### A. Participant information. **participantInfo.xlsx** is an Excel spreadsheet that gives information about participants IDs and who is pilot, study, control. You will see that the IDs are not consecutive numbers -- this is because we always gave everyone interested in the study (and receiving participant information sheet) a number and put them on a separate excel sheet with contact details etc (this is a separate, locked file which is not included with this dataset). The initial participants were pilot subjects from our lab, several of whom trained themselves on their own computers and for whom we don't have their training data (until subject 7). We also include in this spreadsheet some info on people excluded because they were unable to do the task, and/or several which have a bad EEG channel problem that were also excluded from further analysis. ### B. Behavioural and EEG data. There are 4 folders that contain the data. 1. Preprocessed data: * Behaviour: 1 matlab file for main study and also vertical: has response info (see code) and all stimuli saved in cells (easier and faster to read in than to do separately for each subject) (ruesseler-eeg-analysis: behaviouralAnalysis: function: *read\_in\_behav\_data\_with\_new\_response\_matrix.m*) * **preprocessedData/behaviour/behav\_data\_all\_subjs\_all3.mat** contains behavioural data for 28 subjects who completed the main task; 24 for whom EEG data was successfully acquired and a further 4 participants for whom there were technical issues with the data acquisition. The naming conventions of the different variables provided in these two data files are given in ***Appendix A***. * **preprocessedData/behaviour/behav\_data\_all\_subjs\_allVertical.mat** contains behavioural data for the 6 subjects who completed the task with simultaneous vertical and horizontal dot motion stimuli that were superimposed on one another. The naming conventions for these variables are similarly given in ***Appendix A***. * EEG: Folder for each subject with spm files containing raw eeg data for each session for LMRM = left mastoid/right mastoid as reference, downsampled to 100Hz. There are additional preprocessing steps that can be run. (ruesseler-eeg-analysis preprocessing functions: *EyeblinkCopy\_of\_run\_preprocessing\_for\_all\_subjects*) * **preprocessedData/[subjID]/LMRM/Mdspmeeg\_[subjID]\_sess00X\_eeg.mat** contains the EEG data for subject **[subjID]** for session **[X]** (of 6 sessions). This file is in SPM12 format, and has a partner **[.dat]** file with the same filename (all SPM12 files consist of a header .mat file and a data .dat file). The EEG data in this file is after minimal pre-processing has been performed (conversion to SPM12 format, downsampling to 100Hz, and re-referencing of the data using *spm\_eeg\_montage* to a left mastoid/right mastoid reference (hence **[LMRM]** is the folder name)). If you want to perform any analysis with the raw data, we suggest that you use this file (virtually any analysis can be performed with this file, except looking at high-frequency EEG responses). Note that the naming convention Mdspmeeg\_ refers to the fact that the file is in SPM format, and has been downsampled and montaged. Details of the structure of these .mat and .dat files can be understood by referencing the SPM12 manual, available at [https://www.fil.ion.ucl.ac.uk/spm/](https://www.fil.ion.ucl.ac.uk/spm/) * Please note: we also have stored versions of SPM12 files from the different stages of preprocessing, *en route* from the raw downsampled EEG data to the results from the convolutional GLM analyses (see next point). These intermediate preprocessing files are quite large (\~193GB), and so are ***not*** included in this repository. It should be possible to recreate them with the code and the methods from the paper, but please contact us directly if you think you need any of our copies of the intermediate preprocessing stages of the data. 2. convGLM * **convGLM/matchedEegData/LMRM/[subjID]\_EEGdat.mat** are files in MATLAB format, with EEG data matched to stimulus stream (for processing information, look in ruesseler-eeg-analysis repository for: *run\_match\_of\_eeg\_and\_stimulus\_for\_all\_subjects.m*). Note that these include two datafiles for each subject, both before and after current source density analysis (CSD) - the latter is **[subjID]\_csdEEGdat.mat**. The naming conventions for these variables are similarly given in ***Appendix B***. * **convGLM/betasGLMData/LMRM/[subjID]\_[model\_name].mat** are files in MATLAB format that contain the betas for each subject for a range of different convolutional GLMs that have been applied to the data (for details of the different models, look in ruesseler-eeg-analysis for: *run\_convolutional\_GLM\_for\_all\_subjects.m*). The naming conventions for these variables are similarly given in ***Appendix C***. 3. conventionalEEG analysis * **conventionalEEGAnalysis/LMRM/[epoching\_type]\_[subjID].mat** EEG data that have been epoched into trials, either locked to trial start or response, and with/without CSD transform applied (for the different types of epoching, look in ruesseler-eeg-analysis for: *all\_subjects\_create\_single\_trial\_data.m*). The naming conventions for these variables are similarly given in ***Appendix D***. * Note that the folder *averageReference* is intentionally empty, as we only include data that has been re-referenced to a left mastoid/right mastoid reference in this repository (hence the 'LMRM' naming convention throughout). 4. Raw data -- ***removed from this repository.*** * Please note: the *raw* data before conversion to SPM format have been removed from this repository, as Dryad has an upper limit of 300GB per repository. Most analyses should instead be possible with the "lightly preprocessed" data (see point 1), and certainly all analyses that were included in the paper. Please contact us directly if you think you need access to the *raw* data (\~560GB). * Training: folder for each subject with stim and behavioural folders * Experiment: folder for each subject and subfolders for stim (MATLAB format), behaviour (MATLAB), eeg (Curry format), eye (Eyelink) **Maria Ruesseler ([maria.ruesseler@gmail.com](mailto:maria.ruesseler@gmail.com)) and Laurence Hunt ([laurence.hunt@psy.ox.ac.uk](mailto:laurence.hunt@psy.ox.ac.uk))** *** ## Appendix A: Structure of behavioural data files The following variables are stored in the matlab files: **preprocessedData/behaviour/behav\_data\_all\_subjs\_all3.mat** and **preprocessedData/behaviour/behav\_data\_all\_subjs\_allVertical.mat** ### "All\_responses" variable *all\_responses* is a large matrix of *all responses made by allsubjects* \- useful for plotting psychometric functions etc: 1 = points won on current trial or for current response (correct = +3, incorrect = -3, false alarm = -1.5, missed trial = -1.5) 2 = reaction time in secs 3 = choice: 0 left, 1 right 4 = current (mean) coherence of dots 5 = choice: correct 1, incorrect 0 6 = frame on which response occured (for *missed* responses: 500ms after the end of response period) 7 = flag for *type* of response: * 0: incorrect response during coherent motion, * 1: for correct response during coherent motion, * 2: response during incoherent motion, * 3: missed response to coherent motion 8 = total number of trials per block 9 = block ID (which of 4 blocks was currently being completed) * 1: frequent and short response periods * 2: frequent and long response periods * 3: rare and short response periods * 4: rare and long response periods 10 = session id (which of 6 sessions (each session consisted of 4 blocks)) 11 = subject number (running number, starting with 1, used for indexing) 12 = subject ID. Corresponds to *subject ID* column in **participantInfo.xlsx** ### "Streams" variables Variables ending in *streams* are cell arrays indexed with {*subject,session*}(:,*block*). (Note that *subject* here = running subject number (i.e. column 11 in *all\_responses*)) * *stim\_streams:* the actual coherence that participants saw on each frame * *stim\_streams\_org:* what was created before the experimental session-- compare to *stim\_streams*, below * *stim\_streams*: trial periods have been filled with noise after a response is made and the participant detected a trial (and thus contain the actual coherence the participant was exposed to) * *mean\_stim\_streams:* has the mean coherence with which the stimulus was created (0 for baseline periods, 0.3,0.4,0.5 for trials) * *mean\_stim\_streams\_org:* same distinction as for *stim\_streams* * *noise\_streams:* a stream of coherences generated with a mean coherence of 0 over the whole block (used to fill up remaining trial time during the experiment) * *trigger\_streams:* stream of triggers that were attempted to be sent at each frame for EEG recording (not relevant for behavioural analysis) n.b. for all *streams* variables (except *trigger\_streams*): values >1 are set to 1 and values <-1 are set to -1 (as motion coherence can't be greater(smaller) than 1 (-1). Note that stimulus was presented at 100Hz, in blocks of 5min --> \~30,000 frames/block. *** ## Appendix B: Structure of files with EEG data matched to stimulus stream The following two variables are stored in the files **convGLM/matchedEegData/LMRM/[subjID]\_EEGdat.mat:** * *EEGdat{sessionID}{blockID}*: each of these cells contains a channels\*time matrix of EEG data, after the triggers have been matched so that the EEG data corresponds to the same samples in the stimulus streams * *badSamples{sessionID}{blockID}*: each of these cells contains a channels\*time logical index of whether the channel is marked as **bad** in preprocessing (1) or not (0) *** ## Appendix C: Structure of files with betas from convolutional GLM The following variables are stored in the files **convGLM/betasGLMData/LMRM/[subjID]\_[model\_name].mat:** * betas\_subject{regressorNumber}: each of these cells contains a four-dimensional matrix with regression coefficients from the corresponding model fit: dimensions nChannels\*nPeriEventTimepoints\*nSessions\*nBlocks. The ordering of the betas is given by *continuous\_RDK\_set\_options*, which is called by *run\_convolutional\_GLM\_for\_all\_subjects.m* in *ruesseler-eeg-analysis* * chanlabels{channelNumber}: each of these cells contains the label for each of the EEG channels, using 10-20 convention *** ## Appendix D: Structure of files with conventionally epoched EEG data The following variables are stored in the files **conventionalEEGAnalysis/LMRM/[epoching\_type]\_[subjID].mat** * dataAppend: a structure containing epoched data, stored in *FieldTrip* format, for exploratory epoch-based analyses. For details of FieldTrip format visit https://www.fieldtriptoolbox.org |
| Type Of Material | Database/Collection of data |
| Year Produced | 2022 |
| Provided To Others? | Yes |
| URL | https://datadryad.org/stash/dataset/doi:10.5061/dryad.02v6wwq6b |
| Title | Human behavioural and eye-tracking data from inference task with targeted memory reactivation |
| Description | Raw behavioural data is available for 32 participants in .mat (MATLAB) files. Raw eye-tracking data is available for 30 participants in .edf (EyeLink Data file) and .mat formats. Preprocessed data for 27 (behavioural) and 25 (eye-tracking) participants is available in .mat files. Eye-tracking data includes gaze position and pupillometry data for both eyes recorded at 1000Hz. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| URL | https://data.mrc.ox.ac.uk/data-set/human-behavioural-and-eye-tracking-data-inference-task-targeted-m... |
