The broadband dielectric spectroscopy of epidermis at inner forehand using the coaxial probe electrodes with various measurement depth was performed to evaluate the distribution of free water in depth direction and the water relaxation dynamics. The skin at inner forehand was moistened for 30min and was exposed under the environment of 25°C - R.H.20%, followed by the dielectric spectroscopy at few min intervals. It was found that at the deep part of the skin a dielectric relaxation was observed between 100nsec and 1µsec（～10psec for free water）. The distribution of free water volume ratio linearly increases from the surface of the skin to 200µm in depth. At the depth 200 ～600µm, the inclination of the water distribution becomes relatively flat where a small peak exists just after the humidification. The water distribution below 600µm is almost flat at any time.

View full abstract

Recently, in vitro microvasculature system has been greatly demanded in the areas of regenerative medicine and drug development. To construct microvasculature, we seeded human umbilical endothelial cells into the collagen gel channel fabricated by using microneedle-molding method．We examined development of cell stress fibers and cell alignment of the microvasculatures under shear stress．

View full abstract

Increased vascular permeability is a hallmark of inflammation accompanying tumor progression and diabetic retinopathy. For applying to anti-inflammatory drug testing and screening, we have newly developed an in vitro 3D microvasculature model to evaluate vascular permeability. In this model, we could measure and calculate extravasation pattern of FITCdextran fluorescence and determine endothelial barrier changes in response to thrombin, a known pro-inflammatory protease.

View full abstract

In this paper, we focus on integrated approaches to liver tissue equivalent that can be finally used as substitute for those damaged original liver organ. To achieve the large implantable liver, a tissue construct of 500 cm³ in volume with in vivo comparable per-volume-based functionalities is necessary. This could hardly be realized by current technologies due to insufficient mass transfer problems in in vitro engineered tissues. We established a novel engineering methodology by integration of both bottom-up and top-down tissue engineering technologies, which provided an efficient way of arranging engineered liver tissue with improved mass transfer simultaneously from micro- and macro-scales, meanwhile addressing to a clinically relevant size with well recovered per-volume-based functions.

View full abstract

The accuracy of initial guess is important factor to get the solution of Canonical Molecular Orbital （CMO）calculation for large molecules, such as proteins. Quasi-Canonical Localized Orbital （QCLO） method is one of the methods to make precise initial guess of a huge molecule. The previous QCLO program was developed based on peptide structure, so it was difficult to apply it to hetero molecules. Then, we have developed the new QCLO program, which can treat an arbitrary chemical structure, and made it possible to apply QCLO method to hetero molecules. In this study, using the new QCLO program, we analyzed the accuracy of its initial guess for Glucose Oxidase （GOX） model, which has a hetero molecule FAD as its active center. Electronic structure of the GOX active center, which is obtained by CMO calculation, was also studied.

View full abstract

Mathematical modeling has contributed to quantitative understanding of viral infections. The basic model, T'(t)=－βT(t)V(t), I'(t)=βT(t)V(t)－δI(t) and V'(t)=pI(t)－cV(t), has been analyzed which clinical or experimental data sets for many kinds of virus infections so far. However, in the basic model, we implicitly assume that punctual removal of cells and virions due to experimental sampling in cell cultures is described by an exponentially decay. However, the removal of cells and virions is performed instantaneously in viral infection experiments. Therefore, there might be differences among estimated parameters when we use the basic model or an explicit model including the punctual removal. Here, we constructed a hybrid dynamical model which describes the punctual removal by piecewise continuous function to the basic model. We analyzed time course of experimental data for SHIV-KS661 and SHIV-#64 infection in vitro by the basic model and the hybrid dynamical model, and compared the estimated parameters. Interestingly, we found that these two models give similar parameter estimations, and well capture the experimental virus infections. Our results provide a validation of the exponential decay assumption for the punctual removal in terms of parameter estimations.

View full abstract

We tried to get knowledge about the treatment process by using the reinforcement learning based on the MDP environment model as well as medical records of cardiovascular internal medicine especially two kinds of biochemical values （BNP and creatinine） and medication profiles. It is an important future problem to improve configuration of the reward function.

View full abstract

Recently, learning algorithms of multi-layer models, or deep learning, are intensively studied. On the other hand, there is a method of data analysis called the kernel method which is used for analyzing high dimensional data by calculating the "inner products" between two data. However, a multi-layer kernel model has not been studied very much. The purpose of this study is to propose a new multi-layer model using subspace kernel （a multi-layer kernel machine） and its learning algorithm, and evaluate the performance. As a result, it was shown that higher performance can be achieved than existing methods for some data sets.

View full abstract

The dynamic synapses are known to have short-term synaptic plasticity which can store past neuronal activity’s history for about one second as changes of the synaptic transmission efficiency. Therefore, dynamic synapses may play an important functional role in the information processing dealing with time series signals. However, their role is still unclear. In this study, we evaluated short-term memory ability of fully connected recurrent neural networks （RNNs） with dynamic synapses. As a result, we found that dynamic synapses enhance short-term memory ability of the recurrent neural networks. This result suggests that dynamic synapses play an important functional role in short-term memory.

View full abstract

A wide variety of neuron models are studied at different positions in the trade-off between reproducibility of neural activity and computational efficiency. The DSSN model is one of them, which is designed so that it can be effectively implemented in digital arithmetic circuits with small hardware resources. We found appropriate parameter sets for the DSSN model that correspond to the Hodgkin-Huxley-type neuron models of 4 classes of cortical and thalamic neurons. Firstly, we developed 3-variable neuron models by applying a dimension reduction technique to those Hodgkin-Huxley-type neuron models, whose mathematical structures are elucidated by bifurcation analysis. Then, we determined the parameter sets that reproduce these mathematical structures and are able to reproduce characteristic behavior in each class.

View full abstract