Published papers

The number of patients with heart failure and related deaths is rapidly increasing worldwide, making it a major problem. Cardiac hypertrophy is a crucial preliminary step in heart failure, but its treatment has not yet been fully successful. In this study, we established a system to evaluate cardiomyocyte hypertrophy using a deep learning-based high-throughput screening system and identified drugs that inhibit it. First, primary cultured cardiomyocytes from neonatal rats were stimulated by both angiotensin II and endothelin-1, and cellular images were captured using a phase-contrast microscope. Subsequently, we used a deep learning model for instance segmentation and established a system to automatically and unbiasedly evaluate the cardiomyocyte size and perimeter. Using this system, we screened 100 FDA-approved drugs library and identified 12 drugs that inhibited cardiomyocyte hypertrophy. We focused on ezetimibe, a cholesterol absorption inhibitor, that inhibited cardiomyocyte hypertrophy in a dose-dependent manner in vitro. Additionally, ezetimibe improved the cardiac dysfunction induced by pressure overload in mice. These results suggest that the deep learning-based system is useful for the evaluation of cardiomyocyte hypertrophy and drug screening, leading to the development of new treatments for heart failure.

自動高速レーザー照射システムにより目的外細胞を除去し、ロット差の極めて少ない心筋細胞を作製するための前段階として、AIによる機械学習を用いて目的細胞と目的外細胞を識別する手法を開発することを目的とした。自動化心筋細胞拍動評価システムを用いて培養皿上で拍動する心筋細胞の動画を効率よく撮影し、拍動方向のベクトル解析により心筋領域を同定することで非心筋細胞との領域の判別が可能となった。セグメンテーション精度と心筋領域の割合の関係を定性的に確認すると、心筋領域の割合が高い画像のセグメンテーション精度は約80%に達したが、心筋領域の割合が低い画像のセグメンテーションは約1%と非常に低いことが確認された。

Driven by the rapid development of bioimaging, image analysis is now attracting much attention as the technology to extract quantitative information from bioimaging data. In this article, we briefly review recent applications of image analysis to bioimaging data, and introduce our ongoing research on the reconstruction of developmental dynamics of mammalian preimplantation embryos from bioimaging data.

Recursive Diagonal Torus (RDT) is a class of interconnection network consisting of recursively overlaid two-dimensional square diagonal tori for massively parallel computers with up to 216 nodes. Connection structures of the RDT vary according to the assignment of upper rank diagonal tori into a node. Although traditional simple assignment called RDT(2, 4, 1)/α shows enough performance under the uniform traffic, the congestion of low rank tori degrades the performance when local communication is dominant. In this paper, RDT(2, 4, 1)/β torus assignment is proposed, focusing on improving the performance for local communication. With a simplified simulation algorithm, results shows that RDT(2, 4, 1)/β improves the average distance compared with RDT(2, 4, 1)/α assignment when considering local area.

Stochastic biochemical simulation algorithms (SSAs) are generally known as exact methods to trace stochastic behaviors of target biochemical models. Due to vast amount of computation attributed to the nature of Monte Carlo Method, which SSAs are originated from, there is a strong urge for high-throughput execution environment. This paper proposes an FPGA implementation of a stochastic simulation system based on a computationally-efficient SSA called the Next Reaction Method, and studies the evaluation results of area and throughput in detail. The system conducts high-throughput multi-thread execution, using multiple thread modules accessing shared arithmetic and data modules. The network between modules are configurable, and supports flexible network structure according to target FPGAs. In order to evaluate the proposed design, the stochastic simulation system, which is capable of running 16 threads in parallel, was implemented on a middle-range FPGA. As the result of comparing the throughput in RTL simulation with software simulation run on Core 2 Quad Q6600, the system marked 4.2 times higher throughput using a real biochemical model called HSR. When several versions of virtually largescale models were tested on the same simulation environment, maximum of 5.4 times higher throughput was confirmed.

This paper discusses an FPGA implementation and evaluation of a Stochastic Simulation Algorithm (SSA) called the First Reaction Method. SSAs are widely known as rigorous methods for simulating the stochastic behaviors of various biochemical systems, but also as CPU-hogging applications due to vast repetition of the algorithm. This work accelerates the execution by achieving high throughput with concurrent simulations of highly utilized pipelined arithmetic units. For upgrading practical utility, the design stores intermediate data in a BlockRAM so that reconfiguration is unnecessary for different target biochemical systems. Benchmark results on an FPGA (Xilinx XC2VP70-5) have shown that the circuit provides throughput of approximately 83 times and 95 times compared to software execution on Xeon 2.80GHz when it was evaluated with biochemical models called TIS and D4S, respectively.

© 2018 The Authors. Published by FEBS Press and John Wiley & Sons Ltd. Proper determination of the cell division axis is essential during development. Wnt3a is a known regulator of the cell division axis; however, the sensitivity of cells to Wnt3a signalling and its role in determining the cell division axis have not been measured to date. To address this gap, we took advantage of the asymmetric distribution of outer dense fibre 2 (ODF2/cenexin) proteins on centrosomes in dividing cells. To precisely quantify the sensitivity of cells to Wnt3a signalling, we developed a microfluidic cell culture device, which can produce a quantitative gradient of signalling molecules. We confirmed that mitotic SH-SY5Y neuroblastoma cells could detect a 2.5 ~ 5 × 10−3 nm·μm−1 Wnt3a concentration gradient and demonstrated that this gradient is sufficient to affect the determination of the pole-to-pole axis of cell division during the later stages of mitosis.

A reconfigurable biochemical simulator by solving ordinary differential equations has received attention as a personal high speed environment for biochemical researchers. For efficient use of the reconfigurable hardware, static scheduling of high-throughput arithmetic pipeline structures is essential. This paper shows and compares some scheduling alternatives, and analyzes the tradeoffs between performance and hardware amount. Through the evaluation, it is shown that the sharing first scheduling reduces the hardware cost by 33.8% in average, with the up to 11.5% throughput degradation. Effects of sharing of rate law functions are also analyzed. © 2005 IEEE.

細胞内部の生化学反応を数理モデルとして記述し,シミュレーションを行うことで生命現象をシステムとして理解しようとする試みが盛んになっている.しかしながら,汎用のマイクロプロセッサで得られる処理能力は決して十分とはいえず,シミュレーションを高速に実行できる計算基盤が求められている.本論文では,FPGAを用いて常微分方程式を用いた生化学モデルのシミュレーションを高速に実行するシステムであるReCSiPの構成とハードウェアの実装について述べる.評価の結果,最大スループットはXilinx社のFPGAであるXC2VP70-5を用いて,Intel Pentium4 3.2GHzの50〜80倍のスループットを達成することができた.