Organization Chart
Transformative Research Innovation Platform of RIKEN platforms
AGIS: Advanced General Intelligence for Science Program
Program Director: Makoto Taiji
Foundation models of life and medical science project
Program Director Makoto Taiji
Omics AI Research Team
Foundation models in materials science project
Project Director: Ryo Yoshida
Polymeromics Team
Common Infrastructure and Technology project
Project Director: Koichi Takahashi
Universal Foundation Model for Scientific Research Team
"AI for Science" supercomputing platform project
Project Director: Satoshi Matsuoka
Overview
Foundation models of life and medical science project
Project Director: Makoto Taiji
We will develop scientific research foundation models for a comprehensive understanding of living systems across molecular, cellular, tissue, and whole-organism scale layers. Furthermore, we aim to contribute to the medical sciences through models utilizing medical data. To enable the integrated utilization of models across these fields, we will also develop AI agents.
Molecular foundation models for biomedical applications
Theme Leader: Ryosuke Kojima
Our multimodal AI team for molecular science and drug discovery advances cutting-edge AI technologies from an AI for Science perspective, integrating diverse modalities and hierarchical structures across the life sciences. We focus on enhancing large-scale foundation models for diverse types of data, while pursuing research on underlying theories and methodologies. By implementing these technologies and releasing them as tools and platforms, we aim to address a wide range of challenges in the life sciences, including medicine and drug discovery, and accelerate their translation into real-world applications.
Precision Genomic Medicine
Theme Leader: Gen Tamiya
We are developing an AI for Science platform that extends large language models (LLMs) using medical, literature, and electronic health record data to enable the extraction and structuring of critical medical knowledge. The platform incorporates a newly developed genomic language model for medicine to support the simultaneous learning of sequence analysis and clinical data. Furthermore, we develop AI agents that perform everything from variant interpretation to determine pathogenicity to genetic counseling, achieving both advanced analysis and acceptance support.
Artificial Intelligence for Omics
Theme Leader: Itoshi Nikaido
We aim to establish Generative Omics by developing foundational models (AI for Omics) that accelerate life science research. To achieve this, we are pioneering novel genomic methodologies that enable the production of high-quality omics data, capturing diverse cellular responses. Leveraging these datasets, we advance AI for Omics to further enhance its capabilities. Through these technologies, we seek to enable the precise design of cellular functions, drive innovations in drug discovery and regenerative medicine, and contribute to the advancement of next-generation healthcare and the biotechnology industry.
Spatial multicellular foundation models for disease mechanisms
Theme Leader: Yoichiro Yamamoto
Spatial omics technologies that capture cell–cell interactions and disease-specific tissue organization enable insights into disease biology that cannot be obtained from molecular expression levels alone. In this project, we will integrate clinical information from human diseases, cellular imaging data, and spatial omics datasets to construct a broadly applicable foundation model based on human disease spatial information through large-scale pretraining. Through this model, we aim to accelerate the discovery of novel biomarkers, the elucidation of disease mechanisms, and the identification of therapeutic target candidates.
Behavioral and Trait Modeling in Animals
Theme Leader: Hiroshi Masuya
We aim to develop a foundation model that focuses on the relationship between biological individuals and time. We will collect diverse data—including video recordings of daily behavior, audio, and electroencephalogram signals—from mice and marmosets, and apply multimodal machine learning capable of handling spatiotemporal information. By modeling the long-term interactions between objects and behaviors, we aim to advance the understanding of life-stage transitions, early signs of disease and pre-symptomatic states, as well as the development of brain function and communication abilities, and the dynamics of social states.
Life Science Foundation Model Collaboration
Theme Leader: Haruka Ozaki
This theme aims to establish the implementation foundation for scientific research AI agents within AGIS by developing and integrating core technologies, including supervisor agents, specialized agents, MCP server clusters, and laboratory management AI. In parallel, it will establish coordinated integration with foundation models in the life and medical sciences, connecting foundational data, research logs, and experimental operations. Through this integration, the project will construct an autonomous scientific research platform that consistently supports the entire research cycle—from hypothesis generation and experimental execution to analysis and knowledge updating.
Foundation models in materials science project
Project Director: Ryo Yoshida
This project aims to advance materials property research by promoting the development of foundation models for scientific research under the vision of transforming science through AI for Science. By developing and enhancing materials databases, building AI-driven foundation models, and integrating them with automated experimental technologies, we seek to enable data-driven materials discovery. Through these efforts, we aim to promote the autonomous execution of research processes and establish the next generation of scientific research infrastructure.
Advanced General Intelligence for Solid-State Science
Theme Leader: Takahisa Arima
We aim to develop a foundation model capable of proposing candidate materials and their synthesis processes to realize diverse material functionalities. Using magnetic materials as the initial target, we will develop technologies to collect high-quality experimental and computational data. Furthermore, we will promote collaborations with research institutions specializing in materials data, materials DX-related initiatives, and advanced research facilities such as the Fugaku supercomputer and SPring-8. By fine-tuning the foundation model with the collected data, we will develop domain-specific generative models tailored to targeted fields.
Foundation model for material processing
Theme Leader: Shintaro Tani
With the advancement of various material processing technologies, including advanced laser processing, the flexibility of processing methods has increased, making parameter optimization increasingly complex. Meanwhile, simulating processes involving spatiotemporal multiscale irreversible phenomena remains challenging. In this theme, we aim to develop highly predictive process models by tightly integrating fully automated experiments with simulations to enable process optimization in a digital space.
Polymeromics
Theme Leader: Ryo Yoshida
We are building an ultra-large-scale materials database that integrates real-world experimental data with computational simulations, and developing foundation models for polymer materials. By leveraging automated computational experiments based on molecular dynamics and first-principles calculations, we aim to establish one of the world’s largest polymer materials databases. Furthermore, we are advancing Sim2Real machine learning to bridge simulations and experiments, alongside AI- and robotics-driven autonomous materials discovery systems that will accelerate next-generation materials research.
Common Infrastructure and Technology project
Project Director: Koichi Takahashi
We will build an autonomous research environment integrating AI and robotics to realize "Science by AI." We develop and operate multi-AI agent coordination technologies that manage the entire process from hypothesis generation to verification, alongside experimental robotics for high-speed, high-quality data acquisition. Through an active learning architecture fusing foundation AI as the "brain" and robots as the "body," we establish a platform where AI autonomously executes experiments and continuously advances research.
Common Laboratory Infrastructure and Intelligence Embodiment
Theme Leader: Koichi Takahashi
We will develop and operate experimental robots as the "body" of scientific AI, collaborating with various organizations to provide high-throughput, high-quality data for training foundation models. By developing an integrated active learning architecture that combines these robots with the foundation AI as the "brain," we will realize an environment where AI executes experiments and advances its learning autonomously and continuously.
Universal Foundation Model for Scientific Research
Theme Leader: Yoshitaka Ushiku
This theme develops common foundation models—including large multimodal models—for Science by AI. We aim to build a multi-AI agent system that collaborates with researchers to autonomously execute research workflows encompassing steps such as literature survey, hypothesis generation, experimental verification, and reporting. While extending this system to diverse disciplines including bio and materials sciences, we seek to realize a foundational system for science capable of trial-and-error through both computational simulation and real-world experimentation.
"AI for Science" supercomputing platform project
Project Director: Satoshi Matsuoka
The “AI for Science” supercomputing platform project is dedicated to establishing a comprehensive platform for AI for Science (AI-driven scientific research). Central to this mission is the advanced integration of the “Fugaku” supercomputer with the AI for Science development supercomputer. Through this coordinated infrastructure, the project promotes research and development efforts focused on generative AI models tailored for scientific applications.
Through these efforts, the project contributes to innovation in scientific research across diverse fields.
Computing environment and operations for AI for Science
Theme Leader: Shinichi Miura
Our mission is to advance AI for Science by building a highly integrated computational environment that connects the AI for Science supercomputer with the Fugaku supercomputer and state-of-the-art experimental facilities. We seek to enable the automated, real-time application, validation, and feedback of inference and training models, which are essential to the development of foundation models for scientific research, while supporting simultaneous use by hundreds of researchers.
Through the establishment of a cutting-edge computing operations framework, we aim to develop efficient and sustainable operational technologies that balance exceptional performance with cost effectiveness, thereby accelerating scientific discovery and fostering large-scale, collaborative research.
System software for AI for Science
Theme Leader: Kento Sato
We conduct research and development on common application interfaces and workflow technologies that integrate foundation models with scientific applications in the fields of life and medical sciences as well as materials and physical sciences. In addition, we develop performance evaluation and analysis methods, enhance AI frameworks, and build system software and data management infrastructures to efficiently support AI training and inference, aiming to establish an integrated computing platform for AI for Science.
Novel computer architectures for AI for science
Theme Leader: Kentaro Sano
In this research, we aims to establish an architecture that delivers both high-performance training and inference capabilities for AI models, including foundation models, as well as the high-performance computing (HPC) capabilities required for AI for Science (AI4S).
We will develop system software such as compilers and evaluate the training and inference performance of AI models, as well as HPC computational performance, through simulation and emulation.
In addition, to enhance and streamline the optimization of architectural design, we will develop a design-support framework that utilizes AI and related technologies.