TOP ＞ 一般口演（若手道場）

一般口演（若手道場） 若手道場　シュミレーション系 Wakate Dojo: Simulation Themes 座長：味岡 逸樹（東京医科歯科大学／神奈川県立産業技術総合研究所）・豊泉 太郎（理化学研究所 脳神経科学研究センター） 2022年7月1日　14:00～14:15　沖縄コンベンションセンター 会議場B5～7　第4会場 2WD04a1-01 ヒトの動的感覚運動課題における行動と観測の関係性の推論 Inferring the relationship between actions and observations in a dynamic sensorimotor task in humans *岡本 尚之(1,2)、Mike Taylor(3)、Benedetto De Martino(3)、Aurelio Cortese(2) 1. 京都大学情報学研究科、2. 国際電気通信基礎技術研究所脳情報研究所 *Naoyuki Okamoto(1,2), Mike Taylor(3), Benedetto De Martino(3), Aurelio Cortese(2) 1. Grad Sch Informatics, Kyoto Univ, Kyoto, Japan, 2. Computational Neuroscience Labs, ATR Institute International, Kyoto, Japan, 3. Institute for Cognitive Neuroscience, Univ Col London, London, UK Keyword: decision making, sensorimotor integration, rule estimation, computational modeling Humans can behave adaptively in the complicated real world, but much of the underlying mechanisms remain unknown. Oftentimes, humans need to consider whether they are responsible for an observed event, in order to decide if they should update their behavior. For instance, when you press the gas pedal and the car does not accelerate, you will infer the cause (not pressing enough, or the car has an issue) to determine the next course of action. Previous studies have described how agents can infer latent states of the environment from noisy perception, but the ability to infer these latent states from one’s own actions is poorly understood.Here, we aim to evaluate people’s ability to judge whether outcomes are the result of their own actions, or are due to other, random, external causes. We developed a rule-estimating task with a sensorimotor dimension, inspired by the “whack-a-mole” game. In our task, the score shown after participants hit a mole on a touch-screen is calculated by one of two possible latent rules. A “Skill rule” in which the scores reflect how close the hit position is to the center of the mole, and a “Luck rule” in which the scores are calculated randomly. The rule switches happen covertly, at random intervals. After hitting a mole, participants view the score and are instructed to estimate the ongoing rule by taking into account both current and previous scores.Behavior data indicate that participants harbor a specific bias: their performance in the rule estimation tends to be high when the true rule is Skill but low when it is Luck. That is, they are better at recognizing scores generated by their own actions as opposed to scores generated randomly by the environment. In addition, we find that the participants who show overall lower performance seem to have poor sensorimotor skill in hitting the center of the mole, in discriminating the hit location, or a bias in believing that higher scores are the result of their own ability. In order to understand the mechanism of latent rule inference in humans, we construct candidate computational models and use them to simulate new data that mimic participants' choices.Our study provides important insights into how humans weigh self-generated and externally-generated evidence to learn and behave efficiently. 2022年7月1日　14:15～14:30　沖縄コンベンションセンター 会議場B5～7　第4会場 2WD04a1-02 機械学習モデルを応用した新規膠芽腫マーカーの探索 Discovery of a new biomarker of glioblastoma multifome by a machine learning model *岡野 雄士(1)、加瀬 義高(1)、岡野 栄之(1) 1. 慶應義塾大学医学部生理学教室 *Yuji Okano(1), Yoshitaka Kase(1), Hideyuki Okano(1) 1. Dept of Physiol., Keio Univ. Sch. of Med. Keyword: GLIOBLASTOMA MULTIFORME, MACHINE LEARNING MODEL, TRANSCRIPTOME 神経膠腫の一種に分類される膠芽腫は最も高頻度に見られる原発性悪性脳腫瘍である。この疾患は、化学・放射線療法に対して強い抵抗性を示し、再発することが知られている。 　また、膠芽腫患者の 70-80%はイソクエン酸脱水素酵素（IDH）のコーディング遺伝子である、IDH1, IDH2 のいずれかに変異を有しており、このIDHの変異は予後規定因子であることから、膠芽腫および神経膠腫を分類する基準の重要な項目となっている。このように、ゲノムに関する膠芽腫マーカーに関する理解は進んでいるものの、過去の報告における転写産物レベルでの特徴に関しては、様々な遺伝子がマーカーの候補として示唆されているが、結果が解析で用いられるデータセットに大きく依存していることから、再現性が得られておらず、未だ統一した見解が得られていない。 　そこで本研究では、膠芽腫に関する複数のsingle cell RNA-seqデータに対して、膠芽腫の細胞であることを予測する機械学習モデルを構築することによって、一般化可能性について厳密な検討を行いながら、重要な特徴量としての膠芽腫マーカー候補遺伝子を複数発見した。また、既存の研究によって発見された膠芽腫マーカー遺伝子群（HIF1A, VEGFA, CHI3L1, ADGRL4）と比較検討を行った結果、本研究で発見した遺伝子群はこれらマーカーに比して優れた予測性能を有していることが明らかとなった。また、同様に複数のsingle cell RNA-seqデータの解析から、発見された膠芽腫マーカーについて、生物学的観点から、膠芽腫の生存戦略における意義を解析した。 　さらに、臨床経過や病理像などの臨床データが付随する患者検体脳切片を用いて、膠芽腫マーカーの存在を細胞・組織レベルで検討した。また、膠芽腫マーカー候補遺伝子の治療経過に対する影響についても臨床データの解析を行うことで、診断・分類・治療標的化などの観点における有用性について分析および考察を行なったので報告したい。 2022年7月1日　14:30～14:45　沖縄コンベンションセンター 会議場B5～7　第4会場 2WD04a1-03 確率熱力学を用いた脳状態維持コストの定量化 Quantifying the cost of maintaining a brain state based on stochastic thermodynamics *清岡 大毅(1)、大泉 匡史(1) 1. 東京大学大学院総合文化研究科 *Daiki Kiyooka(1), Masafumi Oizumi(1) 1. Grad Sch Arts and Sciences, Univ of Tokyo, Tokyo, Japan Keyword: Stochastic thermodynamics, Brain state maintenance cost, Information theory In our daily lives, we perform various tasks with different levels of cognitive demands. To successfully perform these tasks, the brain needs to control its activity and make its state appropriate for each task. Previous studies have utilized control theory and information theory, and have proposed a framework for quantifying the cost of “transitioning” between different task states. However, few studies have quantified the cost of “maintaining” a brain state for performing a particular task. Here, based on information theory and stochastic thermodynamics, we propose a novel framework to quantify the cost of maintaining a target state given a baseline uncontrolled brain dynamics. To this end, we introduce a quantity which is equipped with clear information-theoretic and stochastic thermodynamic interpretation: minus the change rate of the Kullback-Leibler divergence from a target probability distribution under a baseline dynamics to its equilibrium distribution (Horowitz, Zhou & England, 2017). This quantity not only reflects the intuition that it costs more to maintain a brain state farther from its baseline dynamics but also corresponds to the entropy production, which is a thermodynamic cost. As a proof of concept, we applied the framework to simulation data generated by the Boltzmann machine with asynchronous spin-flop Glauber dynamics. Following the results shown in previous studies, we simulate the awake and the anesthetized brain based on the parameter called temperature. The dynamics and the equilibrium probability distribution are defined as the awake dynamics and the awake state, respectively, when the following conditions are satisfied: the temperature is near a critical point, and three macroscopic quantities (the free energy, the heat capacity and the susceptibility) are high at equilibrium. In contrast, the conditions of the anesthetized state are as follows: the temperature is at a supercritical point, and the three macroscopic quantities are low at equilibrium. We found that it cost more to maintain the awake state when the baseline was the anesthetized dynamics than to maintain the anesthetized state when the baseline was the awake dynamics. This may agree with our intuition that staying awake when we feel sleepy is harder than trying to fall asleep when we do not feel so. The results suggest that our framework serves as a general theoretical tool for investigating the brain state from the perspective of maintenance cost. 2022年7月1日　14:45～15:00　沖縄コンベンションセンター 会議場B5～7　第4会場 2WD04a1-04 社交不安障害(SAD)における小脳の動的脳機能結合 Dynamic connectivity of cerebellum in social anxiety disorder (SAD) *ポポフ ニコライ(1) 1. 国際電気通信基礎技術研究所 *Popov Nikolay(1) 1. ATR Keyword: SOCIAL ANXIETY DISORDER, CEREBELLUM, BRAIN SIGNAL VARIABILITY, BRAIN DYNAMICS Background The role of the cerebellum in social anxiety disorder (SAD) remains unclear. Certain advances have been made expanding our understanding of the role of the cerebellum from purely motor-related functions to social cognition. Functional magnetic resonance imaging (fMRI) has revealed that differences in brain structure are capable of mediating the effect of BOLD signal on behavior, but the understanding of the effect of the cerebellar gray matter (GM) differences on SAD is incomplete. Different fMRI measures, including typical average brain signal and temporal signal variability showed promise in predicting psychiatric treatment outcomes in patients with SAD. Still, the role of the cerebellum was not examined in this context, but might potentially lead to improved neural biomarkers of SAD. Factors such as the cerebellum-specific dynamic states, differences in activation patterns of the cerebellum in SAD and healthy controls (HC) when exposed to naturalistic stimuli were not addressed by previous studies, but could potentially play an important role in predicting anxiety levels and treatment outcomes. Methods Energy Landscape Analysis (ELA) and Hidden Markov Model (HMM) are applied to both rs-fMRI and movie-task fMRI data to extract brain/cerebellum states data. Metrics like traveling score, lingering score for ELA and fractional occupancy, state dwell time for HMM are calculated to analyze the differences between SAD and HC and are used as independent variables in Structural Equation Modeling (SEM) analysis. SEM is used to explore the relationships between behavioral, structural, and functional measurements. To develop a reliable biomarker of SAD, and further explore the possibilities of treatment outcomes prediction that would make use of cerebellar features, naturalistic stimuli-based fMRI is combined with variability-based (moment-to-moment variability) and dynamic connectivity-based models (dynamic mode decomposition). Results Quantified relationships between gray matter volume, fMRI measures and behavioral scores are expected to improve our understanding of the cerebellum in SAD and support the inclusion of the cerebellum in pathophysiological models. Dynamic states of the cerebellum mapped to behavioral measures could be used to track recovery progress, while highly reliable model predicting treatment outcomes would allow to choose best treatment option suitable for a particular patient.