SANTA CLARA — Mitsui & Co., a major Japanese conglomerate, is advancing drug discovery through its new subsidiary, Xeureka, which utilizes NVIDIA’s AI platform to securely harness shared datasets. Founded 77 years ago, Mitsui continues to innovate across its 16 divisions with digital transformation initiatives, ranging from autonomous trucking to geospatial analytics and even quantum computing.

Xeureka, established to drive R&D in healthcare, aims to address one of drug discovery’s most significant challenges: accessing vast datasets needed to train powerful AI models. In drug development, where timelines can extend over a decade and costs reach billions, Xeureka’s goal is to expedite processes while preserving data security. As Katsuya Ito, a project manager in Mitsui’s digital transformation group, explains, “We create businesses using cutting-edge technology like AI and confidential computing,” noting NVIDIA and Fortanix as key technology partners.

Confidential computing plays a crucial role by allowing secure data processing in protected environments within GPUs and CPUs, effectively safeguarding sensitive information. Xeureka recently completed a proof of concept to test this secure approach by simulating two companies with a thousand drug candidates each. Their datasets were used to train individual AI models on chemical toxicity and then combined to build a larger model on NVIDIA H100 Tensor Core GPUs with Fortanix’s security management software. This approach produced a combined AI model with 65-74% greater predictive accuracy, demonstrating the potential of confidential computing for multi-company data collaboration.

The project marked a significant step forward. Xeureka’s CTO, Hiroki Makiguchi, noted, “Confidential computing from NVIDIA and Fortanix alleviates privacy concerns and enhances model accuracy, creating a win-win for the industry.”

Now, Xeureka is expanding this technology within Japan’s healthcare ecosystem through Tokyo-1, an AI supercomputer launched in February to bolster drug discovery research. Tokyo-1’s advanced capabilities allow users to apply large language models to chemical, protein, DNA, and RNA data formats through NVIDIA BioNeMo, a platform designed for drug discovery. Early projects will focus on tasks such as predicting protein structures, screening ligand-base pairs, and enhancing molecular simulations with the supercomputer’s trusted services.

Through Xeureka, Mitsui aims to support Japan’s $100 billion pharmaceutical industry by developing AI-driven services that can screen billions of drug candidates, predict protein interactions, and simulate complex chemical behaviors, solidifying its role as a technological leader in the global pharma landscape.

To learn more, read about NVIDIA Confidential Computing and NVIDIA BioNeMo, an AI platform for drug discovery.