シンポジウム
49 High dimensional neural data: the experimental and theoretical frontiers
座長：Reiter Samuel（Okinawa Institute of Science and Technology Graduate University）・Shi Shoi（University of Tokyo Medical School） |
|---|
| 2022年6月30日 9:00～9:25 ラグナガーデンホテル 羽衣：東 第8会場
1S08m-01
理論駆動の行動課題による知覚意思決定の神経基盤解明
Theory-based behavioral task for understanding neural substrate of perceptual decision making
*船水 章大(1,2)
1. 東京大学定量生命科学研究所、2. 東京大学大学院総合文化研究科
*Akihiro Funamizu(1,2)
1. IQB, University of Tokyo, Tokyo, Japan, 2. Grad Sch Art Sci, University of Tokyo, Tokyo, Japan
Keyword: Perceptual decision making, Reinforcement learning, Neuropixels, Mouse
In perceptual decision-making, prior knowledge of action outcomes is essential to optimize behavior, especially when sensory inputs are insufficient for proper choices. Signal detection theory (SDT) shows that optimal choice bias depends not only on the prior but also the sensory uncertainty. We recently developed a tone-frequency discrimination task for head-fixed mice to investigate how mice integrates the prior knowledge and sensory stimuli with various uncertainties (Funamizu, iscience, 2021). In the task, mice selected either a left or right spout depending on the tone frequency to receive a water reward. We randomly presented either a long or short sound stimulus and biased the outcome for each option. The choice behavior was less accurate and more biased toward the large-reward side in short- than in long-stimulus trials. Analysis with SDT showed that mice did not use a separate, optimal choice threshold in different sound durations. Instead, mice updated one threshold for short and long stimuli with a simple reinforcement-learning rule. During the behavioral task, we are currently recording the neuronal activity of auditory cortex, medial prefrontal cortex, and secondary motor cortex with Neuropixels to investigate how the cerebral cortex integrates the sensory inputs and prior knowledge. Our preliminary result showed that, during the sound presentation, the sound stimuli were encoded in the auditory cortex, while the actions were encoded in the frontal cortices. As our task theoretically dissociates the important factors for optimizing behavior, further neuronal analyses help understanding the neural substrate of perceptual decision making. | | 2022年6月30日 9:25～9:50 ラグナガーデンホテル 羽衣：東 第8会場
1S08m-02
Octopus camouflage: a multiscale approach
*Leenoy Meshulam(1), Aditi Pophale(2), Kazumichi Shimizu(2), Sam Reiter(2)
1. University of Washington, Seattle, USA, 2. Okinawa Institute of Science and Technology Graduate University (OIST), Japan
Keyword: cephalopods, camouflage, entropy, multi-scale
Practice is a crucial component of mastering many movements. However, studies have shown that the rest periods in between practice sessions are equally important. During rest periods, and especially while asleep, human and animal brains undergo a process of ‘offline practice’ resulting in subsequent gains in behavioral performance. Nonetheless, it remains unclear how offline practice produces later improvements. To that end, we study octopus camouflage behavior. Octopuses’ survival is dependent on quickly wielding their skin color and texture to display strikingly complex visual patterns. Here, we leverage the state-of-the-art high-resolution filming setup developed in the Reiter lab at OIST to report a near complete readout of the neural population responsible for generating the collective behavior of skin camouflage patterns. We utilize an entropy-based theoretical framework to model the macro and micro-states of skin patterns, and quantitatively examine the collective nature of skin cell interactions over time. Beyond studying the potentially evolving relationship between patterns adopted during sleep and wake, we are working to show how manipulating the surroundings of the animal while awake affects the content of its practice during sleep. Together, these findings may help elucidate the complex and flexible nature of camouflage and shed light on the more general phenomenon of 'offline practice'. | | 2022年6月30日 9:50～10:05 ラグナガーデンホテル 羽衣：東 第8会場
1S08m-03
大規模解析によるヒト睡眠表現型ランドスケープの可視化
Big data analysis revealed a landscape of human sleep phenotypes
*史 蕭逸(1,2)
1. 筑波大学国際統合睡眠医科学研究機構、2. 東京大学大学院医学系研究科
*Shoi Shi(1,2)
1. International Institute for Integrative Sleep Medicine, Univ of Tsukuba, Japan, 2. Grad Sch Med, Univ of Tokyo, Tokyo, Japan
Keyword: sleep, data analysis, comparative neuroscience
In recent years, the trend of big data analysis has been moving toward a future where in-depth genetic analysis is combined with rich phenotype analysis. There have been many sustained efforts to understand the genetic landscape of diseases such as cancer, neurodegenerative diseases, and psychiatric disorders by conducting large-scale genome-wide association studies (GWAS) and comprehensive analyses of genetic variants. However, in comparison to the depth of genetic analysis, phenotypic analysis still has room for improvement.
Sleep is a physiological phenomenon that is widely conserved throughout the animal kingdom. Its basic structure is genetically conserved within species - humans are no exception to this. However, this structure can change either transiently or chronically depending on environmental factors.
In this study, as a first effort at automatic classification of sleep phenotypes, we classified sleep phenotypes based on over 100,000 accelerometer-acquired dataset in the UK Biobank. The large-scale acceleration data were converted to sleep/wake time series data by combining the state-of-the-art sleep/wake classification algorithm and a non-wear detection algorithm. We calculated 21 sleep indexes from sleep/wake time series data and applied manifold-based dimension reduction and clustering methods. The systematic and unsupervised clustering of large-scale dataset revealed 16 clusters, representing distinctive sleep phenotypes that are consistent with medically described conditions, including social jet lag-related sleep phenotypes and several insomnia-related sleep phenotypes. | | 2022年6月30日 10:05～10:30 ラグナガーデンホテル 羽衣：東 第8会場
1S08m-04
The neuronal dynamics for movement initiation
*Hidehiko Inagaki Inagaki(1)
1. Max Planck Florida Institute for Neuroscience, FL, USA
Keyword: Movement initation, Motor planning, Motor cortex, State space
Motor behaviors are often planned long before execution, but only released after specific sensory events. Planning and execution are each associated with distinct patterns of motor cortex activity. Key questions are how these dynamic activity patterns are generated and how they relate to behavior. Here we mapped the pathways and mechanisms underlying cue-triggered switching of activity modes and the resulting movement initiation, in the context of a delayed directional licking task in mice. By combining anatomy and large-scale, multi-regional electrophysiological recordings, we established the flow of Go cue-related information with millisecond time resolution. Ascending glutamatergic neurons in the midbrain reticular and pedunculopontine nuclei show short-latency and phasic changes in spike rate that are selective for the go cue. This signal is transmitted via the thalamus to the motor cortex, where it triggers a rapid reorganization of motor cortex state from planning-related activity to a motor command, which in turn drives appropriate movement. Altogether, we have identified a multi-regional pathway mediating cue-triggered mode switching and thereby releases planned movements. | | 2022年6月30日 10:30～10:55 ラグナガーデンホテル 羽衣：東 第8会場
1S08m-05
Capturing the evolution of low-dimensional dynamics in large scale neural recordings with sliceTCA
*Alex Alex Cayco Gajic(1), Arthur Pellegrino(2,1), Heike Stein(1)
1. Group for Neural Theory, Departement D'Etudes Cognitives, Ecole Normale Superieure, Paris, France, 2. School of Informatics, University of Edinburgh, Edinburgh, UK
Keyword: population coding, dimensionality reduction, large-scale neural data
A fundamental question in systems neuroscience is how neural circuits encode sensory, motor, and cognitive variables, and how these representations evolve over slow timescales. Towards this end, recent work has proposed applying tensor decomposition methods to identify low-dimensional latent dynamics that vary in amplitude over trials (tensor component analysis; TCA). However, recent evidence suggests that the slow timescale evolution of latent variables over trials may instead be characterized by a reorganization of neural encoding weights (“representational drift”), or by shifts in temporal dynamics as in classic reinforcement learning paradigms. To address this, we propose a new tensor decomposition (sliceTCA) that identifies a broader class of latent low-dimensional dynamics by allowing multilinear dependencies between neural, temporal, or trial factors. We demonstrate that sliceTCA is able to capture large-scale population recordings in fewer components than TCA, aiding interpretability of latent neural dynamics. | | | |
|
|
