| 高阪 翔 | M, 2回目発表 | 数理情報学 | 池田 和司, 金谷 重彦, 別所 康全 (BS), 作村 諭一(BS) |
|
title:Estimation of key molecules and their inhibition levels to regulate biological cell functions by applying machine learning
abstract:If cell functions can be manipulated artificially, effective applications in the medical field can be expected. One of the main cell manipulations is molecular inhibition by drug treatment. Various types of compounds that inhibit biological molecules already exist. Most compounds act on multiple molecules, and the level of inhibition for each molecule is databased as IC50. In order to amplify or reduce specific cell functions, it is necessary to develop compounds that specifically inhibit the related molecules. The purpose of this study is to estimate how much of which molecules should be inhibited to enhance the target cell function. The algorithm learns observations of cellular responses given by known inhibitors by machine learning and predicts unknown inhibitors (molecular inhibition patterns) that enhance cellular responses. In order to evaluate the validity of the method, we constructed a mathematical model of a biochemical reaction network consisting of many molecules. Using this mathematical model, we inhibited the molecules variously and computed the response. The combination of these inhibition patterns and responses was analyzed by this algorithm. It was shown that the molecular inhibition patterns that enhance the cellular response can be estimated without using knowledge about the mathematical model. This methodology not only contribute to the development of novel compounds, but also provide clues to the system elucidation of almost black box cell functions. language of the presentation:Japanese | |||
| 山本 明日翔 | M, 2回目発表 | 数理情報学 | 池田 和司, 金谷 重彦, 川鍋 一晃(客員）, 森本 淳（客員）, 吉本 潤一郎 |
|
title:Elucidation of brain algorithm in learning conceptual knowledge
abstract:Humans can learn new concepts within few trials. It is thought that the prefrontal cortex (PFC) plays an important role in this ability, but how it does so remains largely unknown. We hypothesize that the PFC integrates low-level feature information into abstract concepts when they are tied to value. To test this hypothesis, we used a novel task requiring concept learning while we recorded human brain activity. In the task, subjects learned the fruit preference of an imaginary character through trial-and-error by maximizing their expected rewards. The character had three features: color (red, green), the direction of the mouth direction (right, left), and stripes orientation (vertical, horizontal). The preferred fruit was decided by a random combination of two features, which changed in each block. We found hallmarks of conceptual knowledge learning when subjects performed the task, and that PFC, and in particular ventromedial prefrontal cortex(vmPFC), was involved. in addition, the strength of the functional connectivity between vmPFC and visual cortex(VC) correlated with learning speed. Finally, our results show that several brain areas have more similar brain activities after learning compared with before learning, but only in vmPFC had a clear separation depending on the combination of features. In conclusion, we suggest that vmPFC creates new concept by integrating the importance of lower level features, depending on the current value of the object granted by the reward obtained as a result of previous actions. language of the presentation:Japanese | |||
| 吉井 誉揮 | M, 2回目発表 | 数理情報学 | 池田 和司, 金谷 重彦, 川鍋 一晃(客員）, 森本 淳（客員）, 吉本 潤一郎 |
|
title:Change of information representation in the brain by learning
abstract::Humans can easily and appropriately respond to new problems using the knowledge acquired in the past.Previous studies suggest that high cognitive functions such as attention, memory, concept formation, consciousness, and metacognition interact with reinforcement learning to narrow the search space, and promote efficient learning.Several previous studies indicate that the neural mechanism implementing this framework is likely composed of loops linking Prefrontal Cortex (PFC), Hippocampal formation (HCP), and Basal Ganglia (BG). Yet, it has not been shown how task-dependent changes are represented in these loops. The development of an approach to estimate functional dimensionality that combine singular value decomposition and cross-validation has allowed identifying brain regions that represent task-dependent modulation signals.We applied this approach to data from a decision-making task where subject’s high-dimensional, unconscious brain activity was used to define task conditions.Our results showed that the functional dimensionality of brain regions such as PFC, HCP, and BG, changed over learning sessions. This indicates that as learning progressed, there were changes in the brain regions carrying the task modulation signals.Furthermore, we applied an approach to investigate multivariate statistical dependence between brain regions for BG and PFC of the same tasks. Interaction between BG and PFC changed at specific time points within the trial during learning.Our results add evidence to the proposal that the PFC, BG, and HCP functional loops are involved in the implementation of high cognitive functions (at least memory, consciousness, metacognition) for efficient learning purposes. language of the presentation: Japanese | |||