biomy Advances Development of “Virtual Cell,” an AI Drug Discovery Platform Simulating the Human Tumor Microenvironment, Through Technical Collaboration with NVIDIA

— From accelerating spatial multi-omics analysis to virtual perturbation simulations powered by AI agents —

biomy Inc. (Head Office: Chuo-ku, Tokyo; President and Representative Director: Teppei Konishi; hereinafter “biomy”) announces that, through technical collaboration with NVIDIA, it is advancing the development of “Virtual Cell,” an AI drug discovery platform aimed at recreating the human tumor microenvironment on computers.

In this initiative, biomy is building spatial multi-omics data that integrates gene expression, protein expression, pathology images, and other information, based on clinical tissue specimens being acquired through joint research with the Japanese Foundation for Cancer Research. Using these data, biomy plans to recreate on AI the human tumor microenvironment in which cancer cells, immune cells, stromal cells, and others interact in complex ways, and to apply this integrated understanding of entire tumor tissues—something that conventional single-modality analyses could not fully capture—to drug target discovery, efficacy prediction, and analysis of resistance mechanisms.

biomy is leveraging NVIDIA AI infrastructure and AI software ecosystem for life sciences to accelerate and scale spatial multi-omics analysis, which has historically been constrained by heavy computational workloads. In part of its analysis workflow, biomy has already confirmed a reduction in processing time of up to 90% compared with conventional CPU-based processing. Going forward, biomy plans to build on this acceleration platform and further develop it toward virtual perturbation simulations that combine the Virtual Cell currently under development with AI agents.

Background of This Initiative

In oncology drug development, a major challenge is that drug targets and drug candidates considered promising in basic research and preclinical studies often fail to demonstrate sufficient efficacy in clinical trials. Behind this lies the “wall of preclinical reproducibility”: conventional 2D cell cultures and animal models cannot sufficiently reproduce the spatial structure, cell-cell interactions, immune responses, and drug resistance mechanisms of the tumor microenvironment found in patients.

biomy aims to address these challenges in preclinical reproducibility through an AI for Science approach, and is developing Virtual Cell as its core platform. Using spatial multi-omics data derived from human clinical tissues, biomy is working to realize a platform that reconstructs the tumor microenvironment on computers and predicts the effects of specific genes, cell populations, and drug interventions on the entire tumor tissue.

Through this, biomy will integratively analyze relationships among cell states, spatial arrangements, cell-cell networks, and drug responses that conventional omics analysis and pathology image analysis could not fully capture, and connect these insights to the creation of drug discovery hypotheses with higher clinical translatability.

Technical Collaboration with NVIDIA

biomy, a member of the NVIDIA Inception program for startups,  is developing Virtual Cell with technical collaboration from NVIDIA. This collaboration spans two major areas: improving the efficiency of foundation model training and building AI agents.

1. Improving the Efficiency of Foundation Model Training

Building Virtual Cell requires processing high-dimensional and large-volume spatial multi-omics data obtained from clinical tissue specimens. In particular, the process of integrating and analyzing different types of data, including gene expression, protein expression, and pathology images, involves a continuous series of computationally intensive multi-step processes.

With support from NVIDIA’s technical team, biomy has introduced NVIDIA CUDA-X for data science  libraries to seamlessly integrate GPU acceleration  into its single-cell analysis workflow, and is advancing optimization of its spatial multi-omics data processing platform.

As a result, in part of biomy’s analysis workflow, the company confirmed a reduction in processing time of up to 90% compared with conventional CPU-based processing. This makes it possible to advance the development of Virtual Cell—integrating more clinical specimens, more layers of molecular information, and higher-resolution spatial information—within practical computation times.

2. Building AI Agents

At the same time, biomy is leveraging NVIDIA BioNeMo Agent Toolkit for life sciences to build AI agents that support drug discovery research. Incorporating large language models such as NVIDIA Nemotron open source models and technologies for agent development, biomy is working to develop AI agents that support researchers’ analytical work and hypothesis generation.

Going forward, biomy envisions gradually expanding the Virtual Cell it is currently building into virtual perturbation simulations powered by AI agents.

Specifically, when Virtual Cell is realized, biomy anticipates that, on this platform, it will be possible to predict in silico the effects of suppressing the expression of specific genes, changes in interactions among cell populations, and responses of the tumor microenvironment to drug interventions, thereby efficiently extracting promising drug targets and biomarker candidates.

Furthermore, by utilizing these AI agent technologies, biomy will also work toward realizing a next-generation AI drug discovery workflow in which researchers can enter hypotheses in natural language and AI supports the design of analysis plans, proposes virtual perturbation conditions, interprets simulation results, and suggests candidates for additional experiments.

biomy envisions developing Virtual Cell beyond a mere data analysis platform into an AI drug discovery platform that supports the generation, validation, and prioritization of drug discovery hypotheses. Through the integration of computational science and experimental science, biomy aims to build a next-generation AI for Science drug discovery platform originating in Japan that supports the creation of drug discovery hypotheses with higher clinical translatability.

Comment

Teppei Konishi, President and Representative Director of biomy Inc., commented as follows.

“One of the challenges in oncology drug discovery is that findings obtained in preclinical studies are not always reproduced in the clinic. By combining spatial multi-omics data derived from human clinical tissues with AI and GPU computing infrastructure, we are developing Virtual Cell with the hope of reconstructing the tumor microenvironment on computers and contributing to the creation of drug discovery hypotheses with higher clinical translatability.

Through technical collaboration with NVIDIA, the acceleration of large-scale spatial data processing, which had previously been a constraint, is progressing, and the development of Virtual Cell is also steadily moving forward. We will work toward the development of the Virtual Cell foundation model and its expansion into virtual perturbation simulations using AI agents, and strive to contribute, even in a small way, to global cancer care and drug development as a next-generation AI drug discovery platform originating in Japan.”

Notes

*The reduction in processing time of up to 90% is based on a comparison between conventional CPU-based processing and GPU-accelerated processing in a specific internal analysis workflow at biomy. The effect varies depending on data size, analysis conditions, and hardware configuration.