Simulating the Tumor Microenvironment
Tokyo-based startup biomy is partnering with NVIDIA to develop “Virtual Cell,” an AI platform designed to simulate human tumor microenvironments. By integrating spatial multi-omics data, the technology aims to solve the problem of poor clinical reproducibility in drug discovery, with early workflows already achieving a 90% reduction in processing time.
Oncology researchers face a persistent frustration: the preclinical reproducibility gap. Drug candidates often show promise in 2D cell cultures or animal models, only to fail during clinical trials. According to biomy, this breakdown occurs because current models cannot replicate the complex spatial structures, immune responses, and drug resistance mechanisms found within a patient’s tumor.
To bridge this divide, biomy is building "Virtual Cell," an AI-driven foundation designed to reconstruct the human tumor microenvironment on a computer.
By integrating gene expression, protein expression, and pathological imaging, the system allows researchers to observe interactions between cancer, immune, and stromal cells. This multi-modality approach provides a level of depth that traditional analysis cannot capture, leading to more robust drug target validation and efficacy prediction.
GPU Acceleration Slashes Processing Time
The technical demands of processing high-dimensional, large-scale spatial multi-omics data are immense. To manage these loads, biomy—a member of the NVIDIA Inception program—has integrated GPU acceleration into its development pipeline, utilizing the NVIDIA CUDA-X for Data Science library.
The performance gains are significant. The company reported a 90% reduction in processing time for parts of its analysis workflow compared to traditional CPU-based systems. This efficiency allows the team to incorporate a larger volume of clinical samples, more molecular layers, and higher-resolution spatial data into the model without facing prohibitive wait times. Note that this 90% reduction is based on specific analysis workflows and may vary depending on data size, conditions, and hardware configuration.
Moving Toward Virtual Perturbation
The company plans to move beyond observation to predict how specific genes, cell populations, or drug interventions affect an entire tumor tissue. This shift toward virtual perturbation simulations—powered by the Virtual Cell foundation and AI agents—is intended to help researchers identify drug targets and analyze resistance mechanisms more effectively.
To support this, biomy is utilizing the NVIDIA BioNeMo Agent Toolkit, incorporating large language models like the NVIDIA Nemotron open-source model. This will eventually allow researchers to input hypotheses using natural language, with the AI assisting in planning analysis, proposing perturbation conditions, interpreting results, and suggesting additional experiments.
Bridging Computational and Experimental Science
Through this integration, biomy aims to build a next-generation “AI for Science” drug discovery platform that contributes to global cancer medical care and pharmaceutical development.
Discover more from Archyworldys
Subscribe to get the latest posts sent to your email.