Mitsui & Co.'s Subsidiary Xeureka Leverages NVIDIA AI Platform to Secure Medical Data and Accelerate Drug Discovery

Mitsui & Co., Ltd. subsidiary leverages the NVIDIA AI platform to enable secure sharing of datasets and accelerate drug discovery using robust models built on these datasets.

Mitsui & Co., Ltd. was established 77 years ago. The company maintains its vitality by building businesses and ecosystems using new technologies such as generative AI and confidential computing.

This Tokyo-headquartered conglomerate, comprising 16 divisions, has undertaken digital transformation across multiple domains, including autonomous trucking services and geospatial analytics platforms. Mitsui & Co. has even partnered with a company at the forefront of quantum computing.

In the medical field, bringing a new drug to market can cost over $1 billion and take up to ten years. To address this, Mitsui & Co. has established a new subsidiary, Xeureka, aimed at accelerating research and development in the healthcare sector.

Katsuya Ito, Project Manager in the Digital Transformation Department at Mitsui & Co., stated, “We leverage new digital technologies such as AI and confidential computing to create business solutions. Much of our work is conducted in collaboration with technology companies; for this project, our partners are NVIDIA and Fortanix, a San Francisco-based security software company.”

Xeureka, founded just three years ago, has already completed a proof of concept, addressing one of the biggest challenges in drug development—obtaining sufficient data.

To accelerate drug development, powerful AI models are required; however, the scale of datasets needed to build such models exceeds what most pharmaceutical companies have on hand. Since data typically contains patients’ private information as well as pharmaceutical companies’ proprietary chemical formulations, sharing data among companies was unthinkable until recently.

Confidential computing processes data within a protected portion of the GPU or CPU. This segment functions like a black box, safeguarding an enterprise’s most critical secrets.

To ensure that their data remains confidential at all times, banks, government agencies, and even advertisers are adopting this technology. A consortium of several of the world’s leading companies has also been formed to support it.

To verify whether confidential computing enables customers to share data securely, Xeureka created two hypothetical companies, each with thousands of candidate drugs. The datasets from the two companies were first used separately to train AI models for predicting chemical toxicity levels, and then these datasets were merged to train a similar but larger AI model.

Xeureka atNVIDIA GPU...tested using Fortanix’s security management software. Fortanix is one of the first startups to support confidential computing.

NVIDIA GPUs support Trusted Execution Environments, leveraging hardware-based engines to ensure and verify the protection of confidential workloads on the GPU without compromising performance. Fortanix software manages data sharing, encryption keys, and the entire workflow.

The results obtained were remarkable. By leveraging a combined dataset, the prediction accuracy of the large language model improved by 65–74%.

Ito stated that models built using a single company’s data suffer from instability and bias, whereas large language models do not exhibit these issues.

“Confidential computing by NVIDIA and Fortanix fundamentally alleviates privacy and security concerns while enhancing model accuracy, representing a win-win for the entire industry,” said Hiroki Makiguchi, Chief Technology Officer of Xeureka, in a Fortanix press release.

Xeureka is currently collaborating with the community behind Tokyo-1, a GPU-accelerated AI supercomputer, to explore the broad application of this technology in drug discovery and development. Launched in February this year, Tokyo-1 aims to enhance the efficiency of pharmaceutical companies in Japan and other countries.

Initial projects may include collaborative protein structure prediction, screening of ligand-base pairs, and accelerating molecular dynamics simulations using trusted services. Tokyo-1 users can accessNVIDIA BioNeMoDrug Discovery Microservices and Frameworks, Leveraging Large Language Models for Chemical, Protein, DNA, and RNA Data Formats.

This is part of Mitsui & Co.’s overall strategic development plan. The plan aims to drive the growth of Japan’s $100 billion pharmaceutical industry by developing medical software and services. Japan is the world’s third-largest pharmaceutical market, after the United States and China.

Xeureka’s services will include using AI to rapidly screen billions of candidate drugs, predict how useful molecules bind to proteins, and simulate detailed chemical behaviors.

For more information, please read the following:

• NVIDIA Confidential Computing

  https://www.nvidia.cn/data-center/solutions/confidential-computing/

  
  

• NVIDIA BioNeMo AI Drug Discovery Platform

  https://www.nvidia.cn/clara/biopharma/