Amgen Revolutionizes Drug Discovery with Generative AI, Significantly Shortening New Drug Development Time

Which Company's Case Study?

This case study focuses on Amgen, one of the world's leading biopharmaceutical companies. Amgen is known for its treatments for diseases such as cancer, severe arthritis, and anemia, and is a pioneer in biological drugs (biopharmaceuticals). Addressing the industry-wide challenge of complex, time-consuming, and costly new drug development processes, Amgen is pioneering innovation through the use of generative AI.

Challenges Amgen Aimed to Solve

Traditional drug discovery, especially the development of biologics, is a highly costly and time-consuming process, requiring the screening of tens of thousands to millions of molecules to find effective candidates. Furthermore, even after discovering promising proteins (such as antibodies), many cases failed to reach commercialization due to challenges like manufacturability and stability. Amgen aimed to accelerate and streamline this long and uncertain process using AI.

How AI Was Used

Amgen adopted NVIDIA's GPU-accelerated drug discovery platform, BioNeMo, and its AI supercomputing service, DGX Cloud. This enabled them to rapidly train large language models (LLMs) for proteins, leveraging their vast proprietary data. These AI models are used to design new therapeutic proteins that bind to specific disease targets and to predict and optimize the properties of existing molecules. Through an iterative process (a generative loop) where AI designs and predictive models evaluate, validation work in wet labs (actual laboratories) is significantly reduced, accelerating the development cycle.

Implementation Effects and Key Takeaways

• ▸By adopting NVIDIA's platform, the time required for custom AI model training has been reduced from three months to several weeks.
• ▸Analysis speed after AI model training has been accelerated by up to 100 times through the use of RAPIDS libraries.
• ▸Amgen has succeeded in designing proteins previously thought impossible with conventional methods, leading to new drug candidates under development.
• ▸Notably, AI is being utilized not merely as a data analysis tool but for 'generative' tasks, designing new molecules from scratch.
• ▸Researchers have reported being able to focus on their primary goal of biological exploration rather than infrastructure development.

What Japanese Companies Can Learn

This case study demonstrates that generative AI can be a powerful tool not only in the pharmaceutical industry but also in highly specialized R&D fields. The key is a company's unique datasets accumulated over many years. By training these datasets on the latest AI platforms, it is possible to solve complex industry-specific challenges and create innovations that cannot be replicated by others. The approach of fine-tuning models with proprietary data, rather than just using general-purpose AI, can be applied in many areas, such as material development in manufacturing and risk model construction in the financial industry.

Today's AI News

Reference Links