In many respects, the Laboratory was already primed to tackle pandemic-related challenges. Rapid response to a disease outbreak is a national security priority, and a crucial aspect of LLNL’s biosecurity mission is developing medical countermeasures—such as novel therapeutics—for new pathogens.

“Countermeasures are necessary tools in combating infectious diseases, and we’re now testing technologies we’ve been building over many years, as well as validating and improving our methodologies in preparation for the future. The pandemic propelled us forward,” Lightstone explains.

Underpinning biosecurity projects is Livermore Computing’s (LC’s) powerful ensemble of high performance computing (HPC) systems that enable complex scientific workflows such as virtual screening of massive chemical datasets using machine-learning (ML) models and physics-based simulations—and include the requisite data analysis, memory capacity, workload management, and file storage resources that accompany such R&D. “We were able to move quickly because we could add more compute resources,” notes computer scientist Sam Ade Jacobs.

Moreover, key biomedical projects in progress were readily transferrable to pandemic R&D. For instance, LLNL with program sponsorship from the National Nuclear Security Administration’s Advanced Simulation and Computing Program and from the National Cancer Institute, has participated in the Accelerating Therapeutics for Opportunities in Medicine (ATOM) consortium since its inception in 2017. The consortium aims to accelerate drug discovery through a data-driven ML modeling pipeline.

“We leveraged ATOM’s computational framework including GMD tools to make predictions about small molecules’ drug potential. We’re using these tools to optimize compounds identified by virtual screening for greater safety and potency against SARS-CoV-2,” states Allen. Additionally, the team used the existing Livermore Big Artificial Neural Network (LBANN) toolkit—an HPC-centric deep-learning software framework—to train their ML models.

Similarly, the Laboratory had been engaged with the American Heart Association’s Center for Accelerated Drug Discovery for two years before the pandemic, developing a database of predicted interactions between small molecules and all human proteins.

“We had a prototype of a queryable data portal and the ability to score the interactions, and we were able to compare computational predictions with experiments,” says Computing’s Bioinformatics group leader Marisa Torres. Ultimately, Lightstone notes, “We pivoted from human proteins to viral proteins with the same technology.”