THE PROBLEM: It typically takes 10 to 15 years and more than $1 billion for a drug discovery to reach clinical approval. The process involves complicated safety and efficacy checks, extensive laboratory testing, several clinical trials, and strict regulatory oversight from federal agencies. Though necessary, these steps extend the time and cost required to produce a commercial drug — ensuring only a small percentage of drugs actually get to market. Speeding up that timeline would improve drug discovery and benefit the general population. 

THE SOLUTION: Led by Frederick National Laboratory (FNL) and Lawrence Livermore National Laboratory (LLNL), the Accelerating Therapeutics for Opportunities in Medicine (ATOM) project was designed to develop advanced machine learning tools to shorten the drug discovery timeline. The Department of Energy (DOE) and National Cancer Institute (NCI) signed a five-year Memorandum of Understanding in June 2021 to conduct collaborative research to advance precision oncology and scientific computing. The powerful partnership took a broad approach to cancer research to improve the efficiency and effectiveness of predictive oncology. Using predictive and generative AI software, models, and educational resources, the ATOM team has succeeded in accelerating drugs from discovery to market. 

THE TECH TRANSFER MECHANISM: NCI is the nation’s primary cancer research organization and the world’s largest funder of cancer research. DOE national laboratories bring elite computing, modeling, simulation, machine learning, and AI expertise to the partnership. Within the two agencies, six national laboratories collaborated with two highly regarded research universities — The University of California, San Francisco (UCSF) and Texas A&M University — and some industry partners. All ATOM creations were developed to abide by FAIR (Findable, Accessible, Interoperable, Reusable) principles and and are open-access, available at computational.cancer.gov. This includes the ATOM Modeling PipeLine (AMPL), a software pipeline for building and sharing models to enhance in silico drug discovery with traceability and reproducibility. 

THE OUTCOMES: Researchers have widely adopted ATOM tools for biomedical research. First, the national laboratories are using the technology in their collaborative efforts, such as Oak Ridge National Laboratory’s Frontier supercomputer, where ATOM software is used to train models for international collaborations. At LLNL, AMPL is being used to build property prediction models to screen small-molecule compounds for drug-like properties. UCSF developed a unique student training program designed to equip the next generation of scientists with a multi-disciplinary background in traditional and cutting-edge approaches to drug discovery. FNL is also using the ATOM Modeling Pipeline for its RAS Initiative, which is aimed at combatting cancers with Ras (rat sarcoma) mutations. 

The ATOM team has helped install ATOM tools at key computing centers in the United States, United Kingdom, Germany, and India, including the Food and Drug Administration, Zuse Institute Berlin HPC Center, and Microsoft Azure. As of October 2024, ATOM software had an average of 2,000 viewers and 200 clones per month. ATOM has released nine core tutorials to support new users, released a new version of the software, shared real-world use cases, and hosted an ATOM hackathon at the University of Delaware. In addition, the ATOM team has trained more than 200 college students on its technology.