biomy Advances Development of AI Drug Discovery Platform 'Virtual Cell' with NVIDIA's Technical Collaboration to Recreate Human Tumor Microenvironment

biomy Inc. (Headquarters: Chuo-ku, Tokyo; President and CEO: Teppei Konishi; hereinafter "biomy") announces the advancement of its AI drug discovery platform 'Virtual Cell,' developed in technical collaboration with NVIDIA, aiming to computationally recreate the human tumor microenvironment.

This initiative leverages clinical tissue specimens obtained through collaboration with the non-profit Foundation for Cancer Research to construct spatial multi-omics data integrating gene expression, protein expression, and histopathological images. Using this data, biomy aims to digitally reconstruct the complex interactions among cancer cells, immune cells, and stromal cells within the human tumor microenvironment. By holistically understanding the entire tumor tissue—beyond what single-modality analyses can capture—the platform will support applications in drug target validation, efficacy prediction, and resistance mechanism analysis.

biomy is utilizing NVIDIA's AI infrastructure and life sciences AI software ecosystem to accelerate and scale up spatial multi-omics analysis, which has previously been limited by high computational demands. The company has already achieved up to a 90% reduction in processing time for certain analysis workflows compared to traditional CPU-based processing. Going forward, biomy plans to build upon this accelerated foundation to realize virtual perturbation simulations by integrating Virtual Cell with AI agents currently under development.

Background of the Initiative

In oncology drug development, a major challenge is that promising drug targets and candidates identified in basic research and preclinical trials often fail to demonstrate sufficient efficacy in clinical trials. This is largely due to the 'preclinical reproducibility gap,' where traditional 2D cell cultures and animal models cannot adequately replicate the spatial structure, cell-cell interactions, immune responses, and drug resistance mechanisms present in human tumor microenvironments.

To address this challenge, biomy is pursuing an AI for Science approach, with Virtual Cell as its core platform. By reconstructing the tumor microenvironment in silico using spatial multi-omics data derived from human clinical tissues, biomy aims to predict the systemic impact of specific genetic alterations, cell population dynamics, and drug interventions on the entire tumor tissue.

This integrated analysis will uncover relationships among cell states, spatial arrangements, intercellular networks, and drug responses—insights previously inaccessible through conventional omics or histopathological analyses—enabling the generation of drug discovery hypotheses with higher clinical translatability.

Technical Collaboration with NVIDIA

As a member of NVIDIA's startup-focused Inception program, biomy is collaborating with NVIDIA on the development of Virtual Cell. This collaboration spans two key areas: 'efficiency in foundational model training' and 'AI agent development.'

1. Efficiency in Foundational Model Training

Developing Virtual Cell requires processing high-dimensional, large-scale spatial multi-omics data obtained from clinical tissue specimens. Integrating and analyzing diverse data types—such as gene expression, protein expression, and histopathological images—generates computationally intensive, multi-stage processing workflows.

With technical support from NVIDIA's engineering team, biomy has introduced the NVIDIA CUDA-X for Data Science library to integrate GPU acceleration into its single-cell analysis workflows, optimizing its spatial multi-omics data processing infrastructure.

As a result, biomy has confirmed up to a 90% reduction in processing time for certain analysis workflows compared to traditional CPU-based processing. This enables the integration of larger numbers of clinical specimens, deeper molecular profiling, and higher-resolution spatial data into Virtual Cell development within practical computational timeframes.

2. AI Agent Development

Concurrently, biomy is leveraging NVIDIA's life sciences-focused NVIDIA BioNeMo Agent Toolkit to develop AI agents that support drug discovery research. By incorporating large language models such as the open-source NVIDIA Nemotron and agent development technologies, biomy is building AI agents to assist researchers in data analysis and hypothesis exploration.

biomy plans to progressively expand Virtual Cell—currently under development—into virtual perturbation simulations powered by AI agents.

Specifically, once Virtual Cell is realized, the platform will enable in silico prediction of tumor microenvironment responses to interventions such as gene expression suppression, changes in cell population interactions, and drug treatments. This will allow efficient identification of promising drug targets and biomarker candidates.

Furthermore, by leveraging AI agent technology, biomy aims to create a next-generation AI-driven drug discovery workflow where researchers can input hypotheses in natural language, and the AI assists in designing analysis plans, proposing virtual perturbation conditions, interpreting simulation results, and suggesting follow-up experiments.

biomy envisions evolving Virtual Cell beyond a mere data analysis platform into a comprehensive AI drug discovery platform that supports hypothesis generation, validation, and prioritization. Through the integration of computational and experimental sciences, biomy aims to build a next-generation, Japan-originated AI for Science drug discovery platform that generates hypotheses with greater clinical translatability.

Comment

Teppei Konishi, President and CEO of biomy Inc., stated:

"One of the key challenges in cancer drug discovery is that findings from preclinical studies do not always translate to clinical success. We believe that by combining spatial multi-omics data derived from human clinical tissues with AI and GPU computing infrastructure, we can reconstruct the tumor microenvironment in silico and generate drug discovery hypotheses with higher clinical translatability. This is why we are developing Virtual Cell.

Through our technical collaboration with NVIDIA, we are making steady progress in accelerating large-scale spatial data processing, which was previously a major bottleneck. We will continue advancing the development of the Virtual Cell foundational model and its expansion into AI agent-driven virtual perturbation simulations, striving to contribute, however modestly, to global cancer medicine and drug development as a next-generation, Japan-originated AI drug discovery platform."

Note

* The 90% maximum reduction in processing time is based on internal biomy analysis workflows, comparing traditional CPU-based processing with GPU-accelerated processing. Actual performance may vary depending on data size, analysis conditions, and hardware configuration.