Bio-Generative AI

第1回バイオ生成AI研究会(2025年6月開催)の様子

Recent advances in deep learning have enabled the modeling of complex patterns and latent relationships embedded within massive datasets, as well as the generation of new data based on the learned structures. This capability, referred to as Generative AI, has rapidly evolved into a framework that probabilistically infers underlying structural features from large-scale data and synthesizes novel data with comparable statistical properties. In the domain of natural language processing, large language models (LLMs) such as GPT, Llama, and LLM-jp—developed by NII— apply deep learning to textual corpora. These models capture statistical and contextual dependencies across vast, unstructured datasets, enabling the generation of coherent text and dialogue grounded in the learned probability distributions.

Extending this principle to biological data has given rise to Bio-Generative AI. Among its most prominent applications, the genome Language Model (gLM) learns the statistical and contextual organization of genomic sequences, thereby enabling the prediction of functional and biological characteristics and their subsequent use in rational sequence design. Such approaches mark a transformative shift in life science—from the passive observation and understanding of biological systems to their active design and control—signaling the advent of a new paradigm in biology.

The Vision of the ALIS Center

The ALIS Center seeks to integrate AI methodologies with life science to catalyze the creation of new knowledge and to transform experimental research frameworks. Central to this mission is the development of gLM as a foundational technology for elucidating and designing the structural and functional principles embedded within biological sequences. Furthermore, ALIS promotes the integration of gLM with other foundational models encompassing image, structural, and numerical modalities, while pursuing interoperability with LLM-jp, to construct a comprehensive multimodal AI infrastructure for life science.

At NIG, the predictions generated by gLM are subjected to experimental validation, and the resulting empirical findings are reintegrated into model training—thereby establishing a cyclical research paradigm in which AI models and experimental systems evolve reciprocally. By advancing this iterative cycle, NIG and ALIS aim to construct an integrated experimental platform uniting AI, bio-resources, and phenotypic analysis, ultimately enabling AI-driven understanding and control of biological phenomena and environmental interactions.

For those interested in the gLM project

The Bio-Generative AI Project holds workshops to share development progress and the latest AI research trends. These workshops are open to anyone interested in bio-generative AI, including researchers and students.
For detailed information, please refer to the following link.

バイオ生成AIプロジェクトのご紹介- Introduction to the Bio-Generative AI Project (In Japanese)

Members

• Ken Kurokawa Director
• Koichi Higashi Assistant Professor
• Shinichi Morishita Project Professor
• Kiyoshi Asai Project Professor
• Masahiro Kasahara Project Associate Professor
• Sho Tsukiyama Project researcher

Access

Office

Toranomon 40 MT Building, 5-13-1 Toranomon, Minato-ku, Tokyo, Japan

Lab

Genome Evolution Laboratory, National Institute of Genetics
Yata 1111, Mishima, Shizuoka