Focused Session

Yoshihiro UESAWA
（Meiji Pharmaceutical University）
"Welcome Address by the Newly Appointed Head of the Special Interest Group on Computational ADMET and Future Prospects"

Keiko OGAWA
（Ritsumeikan University
"From Chemical Structure to Adverse Events: Multi-task Models for Adverse Event Occurrence with Interpretable Deep Learning"

Koichi HANDA
（Toxicology & DMPK Research Department, TEIJIN PHARMA LIMITED）
"Machine learning model for drug discovery in DMPK and TOX"

Hideaki MAMADA
（Japan Tobacco Inc.）
"Construction of prediction models for pharmacokinetic parameters by combining molecular images and molecular descriptors"

Shigeki MITAKU
（Nagoya University, Emeritus Professor）
"Analysis of the SARS-CoV-2 genome sequence using the phase diagram of life"

Tetsuya KOBAYASHI
（Institute of Industrial Science, the University of Tokyo）
"Understanding chemical information processing in biological systems"

Yoshihiro UESAWA
（Department of Medical Molecular Informatics, Meiji Pharmaceutical University）
"Applicability in generative AI research and great changes in future research styles"

Shinya YUKI
（Elix, Inc.）
"AI drug discovery in Elix centered on generative models"

Takahiro IKUSHIMA
（Advanced Mathematical Technology Laboratory Co., Ltd.）
"Is there a limit to LLM (Large Language Model)? How does AGI (Artificial General Intelligenc) emerge? Exploring the future"

Kohtaro YUTA
（In Silico Data, Ltd.）
"Applicability of generative AI to autonomous research"

Hirofumi WATANABE
（WithMetis Co., Ltd.）
"AI and XAI for drug discovery research"

Shoichi ISHIDA
（Yokohama City University）
"Exploring Chemical Reaction Network Through Mixed-Reality Interaction"

Simon BENNIE
（Nanome, Inc.）
"VR latest technologies in Theoretical and Computational Chemistry"

The successful creation and optimization of nanoscale biomolecules, including proteins and small-molecule drugs, is often contingent on an in-depth understanding of their three-dimensional structures. VR/MR platforms offer scientists a uniquely powerful and cooperative tool to explore and understand structures and to obtain a distinct perspective of the nanoscale world. This type of platform paves the way for faster and more effective concept development, better communication of scientific principles, and a range of tools for improving early-stage Hit all the way to structure optimization.

Immersive software in science has now developed to the point that it can easily integrate with industry workflows through flexible Python APIs, enabling easier interfacing with widely used modelling methods without scientists having to learn new programming languages or how to develop within game engines. We will highlight the key facets of current and upcoming immersive hardware and how state-of-the-art spatial computing approaches are being used to accelerate drug discovery, both through improved structural understanding and by creating a new paradigm for interacting with standard biochemical research algorithms. We will discuss the collaborative use of algorithms and the unique potential of accelerating their performance through the power of natural human 3D reasoning and direct molecular control.

Jiro KONDO
（Department of Materials and Life Sciences, Sophia University）
"RNA structural motifs as drug targets"

Hisae TATEISHI-KARIMATA
（Frontier Institute for Biomolecular Engineering Research (FIBER), Konan University）
"Regulation of expression for disease-related genes by noncanonical nucleic acids"

Kaori FUKUZAWA
（Graduate School of Pharmaceutical Sciences, Osaka University）
"Molecular interaction analysis of nucleic acid using the FMO method"

Sadao OTA
（Resarch Center for Advanced Science and Technology, The University of Tokyo）
"Developing learning cytometry technologies"

Yodai TAKEI
（Division of Biology and Biological Engineering, California Institute of Technology）
"High-resolution spatial multi-omics analysis of single nuclei"

Nobuhiko KOJIMA
（Graduate School of Nanobioscience, Yokohama City University）
"Possibility of drug discovery by optical observation of spheroids"

Ibuki KAWAMATA
（Tohoku University）
"Introduction of Molecular Robotics Research Group and this session"

Takuya MABUCHI
（Tohoku University）
"Molecular Simulations of Ion Transport Through Artificial DNA Channels"

Masatake SUGITA
（Tokyo Institute of Technology）
"Development and application of molecular dynamics simulation protocols to elucidate the membrane permeation mechanism of cyclic peptides"

Ryuji KAWANO
（Tokyo University of Agriculture and Technology）
"Construction of artificial membrane transporters with nanopore structure"

Teruki HONMA
（RIKEN）
"Overview of AMED DAIIA and Expectations for Information Security Technology"

Jun Jin CHOONG
（Elix, Inc.）
"Efficient and Scalable Framework for Activity Prediction with kMol"

In recent years, the accelerated design and discovery of pharmaceutical compounds through computational methods have driven the need for fast, scalable and secure frameworks to predict ligand-protein interactions. Such predictions are pivotal in identifying potential drug candidates and optimizing their binding affinity without compromising trade secrets within the pharmaceutical industry. However, with the ever-expanding size of chemical and biological databases, conventional computational approaches face challenges in handling the complexity of these tasks. To address this, we introduce a fast and scalable framework with federated learning capabilities for predicting ligand-protein activities. To the best of our knowledge, this is the only framework capable of performing federated learning for drug discovery end-to-end. This framework is developed in collaboration with researchers from Kyoto University. Our approach capitalizes on state-of-the-art machine learning algorithms, advanced feature engineering, and efficient data processing techniques to allow for a fast and scalable prediction end-to-end. Utilizing the power of parallel processing and distributed computing, our framework is capable of handling large datasets, resulting in significantly reduced computation time while maintaining high prediction accuracy. Computation can be performed on CPU or GPU depending on the nature of the task. Furthermore, with federated learning, one can utilize the framework in conjunction with various other collaborators without the need of exposing private data to others. Central to our framework is the customizability of the framework itself. The framework is designed with flexibility in mind, allowing users to customize configuration of models and build a pipeline that is most suitable for the task at hand. Moreover, the framework is capable of exploiting structural information (graph data), structural features, molecular descriptors and protein features. This multi-modal data representation enables a comprehensive analysis of the intricate relationships between ligands and proteins, capturing both global and local interactions that influence binding affinities. By harnessing deep learning architectures such as graph neural networks and convolutional neural networks, our model effectively learns complex patterns and dependencies from the data, thereby enhancing predictive performance. For further performance gain, one can also perform fine-tuning within the framework itself. For analysis purposes, visuals can be generated to have better understanding of the molecular space.

[1] Elix. https://github.com/elix-tech/kmol, 2023.
[2] Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, andGeorge E. Dahl. Neural Message Passing for Quantum Chemistry. In International Conference on Machine Learning, pages 1263–1272, July 2017.
[3] Thomas N. Kipf and Max Welling. Semi-supervised Classification with Graph Convolutional Networks. In Proceedings on the 5th International Conference on Learning Representations, ICLR ’17, 2017.
[4] Ryosuke Kojima, Shoichi Ishida, Masateru Ohta, Hiroaki Iwata, Teruki Honma, and Yasushi Okuno. kGCN: A graph-based deep learning framework for chemical structures. Journal of Cheminformatics, 12(1):32, May 2020.
[5] Qiang Yang, Yang Liu, Tianjian Chen, and Yongxin Tong. Federated Machine Learning: Concept and Applications. ACM Transactions on Intelligent Systems and Technology, 10(2):12:1–12:19, January 2019.
[6] Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning Deep Features for Discriminative Localization. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2921–2929. IEEE Computer Society, June 2016.

Thierry Gilles HANSER
（Lhasa Limited）
"FLuID, Federated Learning using Information Distillation"

Federated Learning (FL) allows to share knowledge across multiple organisations whilst mitigating the risk of leaking private information between partners. In recent years Federated Learning has been adopted in Molecular Discovery [1] to develop a new generation of improved predictive models. In this presentation we will describe a new Federated Learning paradigm where knowledge is transferred using data annotation and Knowledge Distillation [2]. This new data-driven paradigm called FLuID (Federated Learning using Information Distillation) alleviates many of the limitations of the current popular model-driven approach whilst capturing knowledge in a robust and versatile format. We will demonstrate that knowledge can be shared whilst preserving the privacy of the underlying data using a simple, intuitive, and lightweight method. We will present the application of the method to improve the prediction of secondary pharmacology, discuss recent advances in this approach and describe a visual research platform to explore the potential of the method in the industrial context.

[1] Hanser T. Federated learning for molecular discovery. Current Opinion in Structural Biology. 2023 Apr 1;79:102545.
[2] Hinton G, Vinyals O, Dean J. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531. 2015 Mar 9.

Martijn OLDENHOF
（KU Leuven）
"Industry-Scale Orchestrated Federated Learning for Drug Discovery"

To apply federated learning to drug discovery we developed a novel platform in the context of European Innovative Medicines Initiative (IMI) project MELLODDY (grant n°831472), which was comprised of 10 pharmaceutical companies, academic research labs, large industrial companies and startups. The MELLODDY platform was the first industry-scale platform to enable the creation of a global federated model for drug discovery without sharing the confidential data sets of the individual partners. The federated model was trained on the platform by aggregating the gradients of all contributing partners in a cryptographic, secure way following each training iteration. The platform was deployed on an Amazon Web Services (AWS) multi-account architecture running Kubernetes clusters in private subnets. Organisationally, the roles of the different partners were codified as different rights and permissions on the platform and administrated in a decentralized way. The MELLODDY platform generated new scientific discoveries which will be shortly highlighted.

Toshiki KUROSAWA
（Faculty of Pharma-Science, Teikyo University）
"Transport analysis and application to MPS using a human iPS cell-derived blood-brain barrier model"

Ryuji YOKOKAWA
（Department of Micro Engineering, Kyoto University）
"Microphysiological Systems (MPS) With Vascular Network for Tumor Microenvironment and Organogenesis Applications"

Yamamoto Yuki
（HiLung Inc）
"Application of lung organoids for future innovative therapeutics"

Shigenori TANAKA
（Kobe University）
Introduction

Yoshiharu MORI
（Kobe University）
"Molecular mechanism of initiation tRNA recognition in translation initiation process: Exploring the evolution of tRNAs and ribosomal proteins"

Yuma HANDA
（Hoshi University）
"Analysis of interaction between translation initiation factor eIF4A and RNA"

Ryosuke KOJIMA
（Kyoto University）
"Multi-modal medical-related data analysis based on large-scale graph neural networks"

Kento TOKUYAMA
（CHUGAI PHARMACEUTICAL）
"Practical AI Applications in Drug Discovery Research by Chugai Pharmaceutical"

Satoshi MIZUNO
（Tohoku University）
"The early prediction models of low-birth-weight reveals different influential environmental and genetic factors for prediction between term and preterm birth group"

Kazuki NAKAMURA
（KYOWA HAKKO BIO CO., Ltd.）
"AI technologies for effective treatment and disease prevention at the individual level"

Wataru FUJIBUCHI
（Japan Shinyaku Co., Ltd. / Graduate School of Medicine, The University of Tokyo）
"Features and Its Value of StemPanTox, Human Stem Cell Multi-Organ Toxicity Prediction System"

Toshiro TAKASE
（Japan IBM Corporation）
"Reusable Execution Pipeline Build: Introducing the Stem Pantox α"

Tsuyoshi KATO
（Gunma University）
"Study of chemical toxicity prediction algorithm for StemPantox"