| Project Area | Deciphering and Manipulating Brain Dynamics for Emergence of Behaviour Change in Multidimensional Biology |
|---|
| Project/Area Number |
22H05156
|
|---|
| Research Category |
Grant-in-Aid for Transformative Research Areas (A)
|
| Allocation Type | Single-year Grants |
|---|
| Review Section |
Transformative Research Areas, Section (III)
|
|---|
| Research Institution | Advanced Telecommunications Research Institute International |
|---|
Principal Investigator |
CORTESE Aurelio 株式会社国際電気通信基礎技術研究所, 脳情報通信総合研究所, 研究室長 (60842028)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
川人 光男 株式会社国際電気通信基礎技術研究所, 脳情報通信総合研究所, 所長 (10144445)
細谷 晴夫 株式会社国際電気通信基礎技術研究所, 脳情報通信総合研究所, 主任研究員 (50335296)
|
|---|
| Project Period (FY) |
2022-06-16 – 2027-03-31
|
| Project Status |
Granted (Fiscal Year 2025)
|
| Budget Amount *help |
¥108,420,000 (Direct Cost: ¥83,400,000、Indirect Cost: ¥25,020,000)
Fiscal Year 2025: ¥24,960,000 (Direct Cost: ¥19,200,000、Indirect Cost: ¥5,760,000)
Fiscal Year 2024: ¥24,830,000 (Direct Cost: ¥19,100,000、Indirect Cost: ¥5,730,000)
Fiscal Year 2023: ¥14,170,000 (Direct Cost: ¥10,900,000、Indirect Cost: ¥3,270,000)
Fiscal Year 2022: ¥19,240,000 (Direct Cost: ¥14,800,000、Indirect Cost: ¥4,440,000)
|
|---|
| Keywords | adaptive behavior change / reinforcement learning / metacognition / neural dynamics / cerebellum / Adaptive behavior change / decision-making / prefrontal cortex / neuroimaging / decoded neurofeedback / learning / probabilistic PCA / behavioral change / neurofeedback |
|---|
| Outline of Research at the Start |
Our research aims to revolutionise how we understand behaviour change. Over five years we will develop new computer algorithms that can very accurately analyse many behaviour measures from recordings of neural activity. Our team will apply machine learning techniques to complicated behavioural and neural data acquired from experiments in which mice and humans solve decision-making problems. We will then combine our new algorithms with advanced brain imaging in a novel experiment design, in both humans and mice, to control patterns of brain activity and cause changes in the targeted behaviour.
|
| Outline of Annual Research Achievements |
Research stream 1 (Hoang, Toyama, Kawato, Cortese): Our research has advanced our understanding of Go/No-go data by incorporating a reinforcement learning model (Hoang et al.2023, eLife). Our findings suggest collaboration between the cerebellum, basal ganglia, and cortex in forming a modular reinforcement learning system. This system involves distinct cerebellar modules receiving reward-prediction errors via climbing fibre inputs, enabling the generation of precise motor commands for Go and No-go cues (Hoang et al. 2023, bioRxiv). Additionally, we have refined a method to detect hundreds of individual climbing fibres across various two-photon recording sessions while learning Go/No-go tasks. These dual achievements provide valuable insights and resources for constructing a sophisticated model of the cerebellum.
Research stream 2 (Okamoto, Six, Taylor, Ovadia, Oka, Gutierrez, Lobaskin, Cortese): We collected data with novel behaviour tasks to test behaviour change through various measures. All tasks shared a common structure, in which participants updated their behaviour to sudden hidden changes in task conditions. We collected participants' subjective confidence ratings along multiple dimensions (perceptual, rule confidence) as well as reaction time data. Our analysis (Taylor, Ovadia, Oka, Lobaskin, Cortese) shows we can obtain reliable behavioural measures with confidence in predicting subsequent behaviour and decision-making. The results were presented at international conferences. In addition, we published an influential opinion paper (De Martino and Cortese 2024, TICS).
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Research stream 1 (Hoang, Toyama, Kawato, Cortese): Our current focus lies in developing a model of the cerebellum, which is strongly informed by our recent findings. In our model, these modules are equipped with bidirectional plasticity at the parallel fibre-Purkinje cell synapses. This means climbing fibre inputs convey reward-prediction errors to enhance or decrease Purkinje cell activity, depending on the specific learning objectives. Our preliminary results are promising. The model has successfully replicated the firing activity of climbing fibres observed in the two cerebellar modules, as well as the licking behaviour of mice in Go/No-go tasks. This suggests that our model accurately captures key cerebellar function and behaviour aspects, laying a strong foundation for further exploration and refinement.
Research stream 2 (Okamoto, Six, Taylor, Ovadia, Oka, Gutierrez, Lobaskin, Cortese): Our current focus is characterising multidimensional confidence representations at behaviour, computational, and neural levels. We are writing two manuscripts, which we expect to publish within the current fiscal year, reporting on (i) our study linking learning, behaviour adaptation, and metacognition (Okamoto et al., in prep), and on the neural correlates of metacognitive evaluation (Six et al., in prep). In parallel, we are analysing behaviour and neuroimaging data (fMRI, ECoG, MEG, by Ovadia, Six, Taylor, Lobaskin).
|
| Strategy for Future Research Activity |
Research stream 1 (Hoang, Toyama, Kawato, Cortese): We will construct a large-scale spiking neuron network that includes various neuron types found within the cerebellum. This model will have a modular structure to reflect the organisational principles observed in the cerebellar cortex. The primary focus will be investigating the role of positive feedback loops and Purkinje cells, cerebellar nucleus neurons, and inferior olive neurons in generating self-organising network dynamics. In parallel, we will begin preparation with the Matsuzaki lab on the mouse DecNef project. We will define the target cortical regions and neural population coverage, the decoding approach, and the behaviour dimensions to be modified through neurofeedback.
Research stream 2 (Okamoto, Six, Taylor, Ovadia, Oka, Gutierrez, Lobaskin, Kasahara, Cortese): We will analyse and apply computational models such as reinforcement learning and hierarchical models, to study different confidence signals (perceptual, rule confidence) and their contributions to behaviour strategies. Importantly, we comprehensively evaluate low-dimensional metacognitive representations in time and space across neuroimaging modalities (fMRI, MEG, ECoG). Next, we will collect eye-tracking data to evaluate physiological confidence correlates. We will design and pilot the neurofeedback experiment in the fiscal year's second half.
Finally, we plan to present our findings from both research streams at reputed international conferences.
|
