オンライン・キックオフシンポジウム (3月12日(金))

「AIとVRを活用した分子ロボット共創環境の研究開発」プロジェクト(2020年度NEDO採択テーマ)キックオフシンポジウム・プログラム(3月12日(金))

日時: 2021年3月12日(金)10:00 – 17:00
主催: 合同会社分子ロボット総合研究所
併催: 情報処理学会バイオ情報学研究会(3月11日)、
   オープンバイオ研究会(3月13日)
参加方法 オンラインシンポジウム参加のための情報を送付いたしますので、下記のメールアドレスに名前および所属をご連絡ください。
送付先:  kickoff[at]molecular-robotics.co.jp [at]を@に変えてください。
件名: キックオフシンポジウム参加希望

開催趣旨
コロナ禍で、仮想現実(VR)を用いた共同作業や遠隔操作への期待が更に増大しています。NEDO/VR共創環境プロジェクトでは、AI予測同期制御により、ネットワーク遅延を解消し、自然な遠隔VR操作を実現する技術の開発を目指しています。本シンポジウムでは、NEDOプロジェクトマネージャによる「人と共に進化する次世代人工知能に関する技術開発事業」に関する基調講演に引き続き、共同研究者によるDNAロボットや微小管群ロボットを創製する分子ロボット技術ならびにVR共創環境を実現するための情報処理技術についてご紹介いたします。

プログラム
10:00-10:10 オープニング  小長谷明彦(合同会社分子ロボット総合研究所)
10:10-10:40 基調講演 仙洞田 充プロジェクトマネージャ(NEDOロボット・AI部)
「人と共に進化する次世代人工知能に関する技術開発事業」ご紹介

少子高齢化により労働生産人口が減少する中、労働生産性を向上させ、顕在化する様々な社会課題を解決する手段としてAI 技術は有効です。本事業では、これまで人ではないとできないと考えられていた領域にAI 技術を適用する基盤技術を開発し、AI ならではの判断とその根拠から人が新たな気付きを得られると同時に、人の経験や知識をAI に導入することで人の感性に合ったより身近なAI システムの実現を目指します。


10:40-11:20 招待講演 葛谷明紀(関西大学)
Nanomechanical DNA Origami Devices Suitable for High-Speed AFM

We have developed various nanomechanical DNA origami devices (DNA origami pliers and DNA origami forceps) that can selectively interact with target molecules in single-molecule manner and transform into designated structure. Wide range of target molecules are applicable, such as metal ions (Na+, K+, and Ag+, etc.), small molecules (ATP and biotin), or even huge proteins (streptavidin or IgG), by selecting appropriate ligands such as aptamers, DNA quadruplexes, and mismatched base-pairs. These devices were designed according to the original DNA origami principle, i.e. bundling multiple DNA helices in plane. We recently found that such planar DNA origami structures are not always suitable for high-speed (HS) AFM observation, particularly when large structural change of the molecular devices take place. We report a new nanomechanical DNA origami pinching devices with rigid lever portions designed in Honeycomb-lattice 3D design principle, DNA origami chopsticks and successful real-time HS-AFM imaging of their target-selective structural changes on mica.


11:20-12:00 招待講演 角五彰(北海道大学)
Molecular Swarm Robot in Real Life
Recent advancements in molecular robotics have been greatly contributed by the progress in various fields of science and technology, particularly in supramolecular chemistry, bio- and nanotechnology, and informatics. Yet one of the biggest challenges in molecular robotics has been controlling a large number of robots at a time and employing the robots for any specific task as flocks in order to harness emergent functions. Swarming of molecular robots has emerged as a new paradigm with potentials to overcome this hurdle in molecular robotics. The latest developments in swarm molecular robotics, particularly emphasizing the effective utilization of bio- and nanotechnology in swarming of molecular robots will be discussed. Prospects of the molecular swarm robots in utilizing the emergent functions through swarming are also emphasized together with their future perspectives.
12:00-13:30 ランチタイム

13:30-14:00 ロングトーク Gregory Gutmann (東京工業大学)
VR for Enhanced Perspectives, Knowledge Creation and Guiding Experimentation for Molecular Research

Almost everything in the world around us is made up of molecules from the food we eat, the medicines that we take, the materials that we manufacture, the list is seemingly endless. As such, much of what we interact with day to day has been a product of or could be improved by molecular research. However, despite the role molecules play our life we cannot see them or interact with them naturally to gain intuition on their dynamics. Also, the sheer number of molecules that make up objects of interest make it difficult to simulate in perfect detail, and the problems faced at the molecular scale are often considered NP-hard problems. In an effort to alleviate some of these challenges, we are proposing the use of a coarse-grained, human-in-the-loop VR system for molecular research. The use of coarse-graining enables us to model larger problems at interactive rates for VR, and the hands-on approach in VR enables users of the system to either gain intuition or use their existing intuition to help find a path to their desired result more quickly. The benefits of our proposed system include offering enhanced perspectives, over typical numerical or monitor-based molecular research tools, by allowing users to jump into a virtual molecular world to interact with molecules in a familiar way to their everyday life. Second, our system enables greater knowledge creation and sharing through cloud based collaboration in VR, as communicating molecular research through data alone can be prohibitive. Third, our VR system can be used for guiding experimentation enabling users to augment numerical methods with their own intuition and knowledge.


14:00-14:30 ロングトーク Arif Pramudwiatomoko (東京工業大学)
Tensegrity Representation Model for Viscoelastic Biomolecule 3D Objects in an Interactive Virtual Reality Environment

The mechanical properties of biomolecules play an important role in the emergence of global dynamics in a massive swarm of objects. The tensegrity representation model was developed to create 3D viscoelastic biomolecule objects with wide-range elasticity from very flexible objects to very rigid objects. The tensegrity representation method binds the object particles to the anchors around them with springs to form a tensegrity structure. The tensegrity object shows the shape of the bending in accordance with the classical bending equation. The object exhibits viscoelastic behavior referring to the Kelvin (Voigt) rheological model. The flexural rigidity of the object can be adjusted using the parameter fitting function with a wide range of elasticity without the dependence of particle formation. Combined with a full-hands natural user interface in a virtual reality (VR) system, this model forms a platform for interactive VR molecular simulations.


14:30-15:00 ロングトーク Zhang Yuhui (東京工業大学)
Compressive Auto-Encoder of Point Clouds with Irregular Convolutions
Point clouds are powerful, flexible and accurate representatives of 3D geometries. However, it is generally hard to compress a dense point cloud, even with data loss. This is mainly due to the lack of regularized data layouts in Point Clouds, such as grids. In this work, we propose Irregular Convolutional Neural Networks for Point Clouds, specially designed for such irregularities. We demonstrate the efficiency and adaptability of such models given a particular data domain, such as daily 3D shapes (ShapeNetCore) or particle simulation data. Furthermore, we show that it is even possible to learn and simulate the hidden dynamics of particle simulation, using the compressed objects without reconstruction. We also compare the model with other state-of-the-art Machine Learning models, and several visualizations were done to intuitively understand the model.


15:00-15:30 休憩

15:30-15:50 ショートトークMa Chen (東京工業大学)
Tracking Microtubule Groups on Gliding Assay Videos

Microtubules are often occurred into groups which have similar moving directions in microtubule gliding assay. They usually moves randomly in low density, comes to several bundles when increasing density and tens to moving together. Also they sometimes moves into a circle. Tracking microtubule groups often becomes difficult due to the high density, low signal to noise (SNR), and sudden appearing and disappearing of a microtubule. It is also tedious and time consuming to track them manually because of the large amount and overlapped situations.

This research aims to develop workflow for the group movement of microtubules. And explorer its feasibility with experimental video data followed by performance evaluation on CPU and GPU implementations. Now our workflow consists of the U-Net like Fully Convolutional Neural Network (FCN) for noise filtering, Sparse Optical Flow (SOF) for tracking and SOF cluster for matching between frames. In this workflow, we firstly adjust the parameters using the simulated microtubule gliding assays graphic videos, and then applied on real experiment video data for evaluation. The workflow can drastically accelerate its performance by GPU with CUDA parallel programming.


15:50-16:10 ショートトークHu Xiaoran (東京工業大学)
AFM DNA Image Super Resolution with Deep Learning and 3D Simulation Model
 Atomic force microscopy (AFM) is a high-resolution type of scanning probe microscopy and has the capability of imaging nanoscale structures at solid interface. So, it has been widely used in many fields and it is a common way to visualize the movement of DNA in suitable conditions. However, there are some limitations on imaging resolution, which makes it difficult to observe the details of double helix DNA sequence, like major groove and minor groove. Deep learning is a part of machine learning methods and several deep learning frameworks have been applied in super resolution task and achieved great result. So, this research tries to apply deep learning method to build super resolution image for AFM DNA images. Since it is very difficult to get super resolution AFM images, we attempt to build 3D double helix DNA simulation model which has the same shape of real DNA strands. And we treat the 3D simulation model as the targets to train a super resolution neural network for AFM DNA images to see the details of DNA strands.

16:10-16:30 ショートトーク Hirotaka Kondo (関西大学)
A Manipulator as an Emulator of AFM and its Prototype Operating System in a VR Environment

We will use DNA origami molecular machines to construct DNA robots to manipulate single molecules in a VR environment.

Our nanomechanical DNA origami devices (DNA pliers) developed in the past can bind a single target molecule selectively, and then they take closed-form structures. The DNA pliers are expected to work as grippers for molecular-sized robot arms.

As a goal for five years, we will construct a nanoscale manipulator and its operating system that can be manipulated intuitively in a VR environment. The system manipulates an AFM probe with such a DNA origami device attached to its tip in real-time. AFM probe and DNA origami devices are regarded as a robot arm and end-effectors which works in molecular level resolution.

However, it is microscopic and difficult to understand whether the system is working correctly. So we decided to create a visually easy system at first. We used a dispensing machine as an emulator of the AFM and a Vive Pro Eye as a manipulating VR device. The prototype operating system was constructed with Unreal Engine 4 and the dispenser’s control program.

I will demonstrate the emulator of AFM and the prototype system.


16:30-16:55 講評およびディスカッション
16:55-17:00 クロージング