Established 77 years in the past, Mitsui & Co stays vibrant by constructing companies and ecosystems with new applied sciences like generative AI and confidential computing.
Digital transformation takes many varieties on the Tokyo-based conglomerate with 16 divisions. In a single case, it’s an autonomous trucking service, in one other it’s a geospatial evaluation platform. Mitsui even collaborates with a accomplice at the forefront of quantum computing.
One new subsidiary, Xeureka, goals to speed up R&D in healthcare, the place it may well take greater than a billion {dollars} spent over a decade to convey to market a brand new drug.
“We create companies utilizing new digital know-how like AI and confidential computing,” stated Katsuya Ito, a undertaking supervisor in Mitsui’s digital transformation group. “Most of our work is completed in collaboration with tech corporations — on this case NVIDIA and Fortanix,” a San Francisco primarily based safety software program firm.
In Pursuit of Huge Information
Although solely three years previous, Xeureka already accomplished a proof of idea addressing one in every of drug discovery’s largest issues — getting sufficient information.
Rushing drug discovery requires highly effective AI fashions constructed with datasets bigger than most pharmaceutical corporations have readily available. Till lately, sharing throughout corporations has been unthinkable as a result of information usually accommodates personal affected person data in addition to chemical formulation proprietary to the drug firm.
Enter confidential computing, a means of processing information in a protected a part of a GPU or CPU that acts like a black field for a company’s most necessary secrets and techniques.
To make sure their information is stored confidential always, banks, authorities companies and even advertisers are utilizing the know-how that’s backed by a consortium of a number of the world’s largest corporations.
A Proof of Idea for Privateness
To validate that confidential computing would permit its prospects to soundly share information, Xeureka created two imaginary corporations, every with a thousand drug candidates. Every firm’s dataset was used individually to coach an AI mannequin to foretell the chemical compounds’ toxicity ranges. Then the info was mixed to coach an analogous, however bigger AI mannequin.
Xeureka ran its check on NVIDIA H100 Tensor Core GPUs utilizing safety administration software program from Fortanix, one of many first startups to assist confidential computing.
The H100 GPUs assist a trusted execution surroundings with hardware-based engines that guarantee and validate confidential workloads are protected whereas in use on the GPU, with out compromising efficiency. The Fortanix software program manages information sharing, encryption keys and the general workflow.
As much as 74% Greater Accuracy
The outcomes have been spectacular. The bigger mannequin’s predictions have been 65-74% extra correct, thanks to make use of of the mixed datasets.
The fashions created by a single firm’s information confirmed instability and bias points that weren’t current with the bigger mannequin, Ito stated.
“Confidential computing from NVIDIA and Fortanix primarily alleviates the privateness and safety considerations whereas additionally bettering mannequin accuracy, which is able to show to be a win-win scenario for your complete trade,” stated Xeureka’s CTO, Hiroki Makiguchi, in a Fortanix press launch.
An AI Supercomputing Ecosystem
Now, Xeureka is exploring broad purposes of this know-how in drug discovery analysis, in collaboration with the neighborhood behind Tokyo-1, its GPU-accelerated AI supercomputer. Introduced in February, Tokyo-1 goals to boost the effectivity of pharmaceutical corporations in Japan and past.
Preliminary tasks could embrace collaborations to foretell protein buildings, display ligand-base pairs and speed up molecular dynamics simulations with trusted companies. Tokyo-1 customers can harness massive language fashions for chemistry, protein, DNA and RNA information codecs by way of the NVIDIA BioNeMo drug discovery microservices and framework.
It’s a part of Mitsui’s broader strategic development plan to develop software program and companies for healthcare, resembling powering Japan’s $100 billion pharma trade, the world’s third largest following the U.S. and China.
Xeueka’s companies will embrace utilizing AI to shortly display billions of drug candidates, to foretell how helpful molecules will bind with proteins and to simulate detailed chemical behaviors.
To be taught extra, examine NVIDIA Confidential Computing and NVIDIA BioNeMo, an AI platform for drug discovery.