Program Accomplishments

Livermore’s LDRD program invests in research aimed at meeting some of the nation’s most difficult challenges. Shortly after the COVID-19 pandemic emerged in early 2020, these internally directed funds enabled the Laboratory to rapidly launch new projects and redirect the focus of existing projects to help the nation respond to the pandemic. 

According to Doug Rotman, who leads the LDRD program at Livermore, “We had several scientists contact us with ideas regarding how they could help our Laboratory respond to the national COVID challenge, including emerging needs related to therapeutic targets and designs. Working closely with science leaders across LLNL, the LDRD program rigorously and quickly evaluated proposals and approved funding needed to start work.”

The LDRD-sponsored projects leveraged LLNL’s capabilities in high-performance computing, simulation, and data science—combined with Livermore’s expertise in bioscience and bioengineering—to identify risk factors for COVID-19 patients and accelerate development of medical countermeasures. 

LLNL combines bioinformatics expertise and supercomputing resources to explore antiviral drug design

Soon after the pandemic was recognized as a global crisis, leaders of three LDRD-funded projects quickly expanded the scope of their projects, focusing on efforts to identify new antibodies and antivirals that could effectively treat a COVID-19 infection. 

Armed with the predicted protein structure of especially important parts of the viral replication machinery and a handful of antibodies known to bind and neutralize SARS-CoV-1 (a similar coronavirus that causes Severe Acute Respiratory Syndrome), researchers used LLNL’s high-performance computing platforms and artificial intelligence resources to virtually screen antibodies capable of binding to SARS-CoV-2 virus receptors and neutralizing them. Their novel approach integrated bioinformatic modeling and molecular simulations, driven by a machine-learning algorithm, to identify antibody candidates. 

These techniques provided the team with the speed and scalability needed to search and evaluate huge numbers of possible antibody designs and reduce the possibilities to 20 initial sequences predicted to target SARS-CoV-2. This approach sped up the process considerably, and enabled scientists to focus their search on areas they might otherwise have overlooked—in the appropriate design space.

“Being able to rapidly estimate antibody structures makes it possible to follow up on promising predictions as they emerge,” said LLNL data scientist Tom Desautels, principal investigator for one of the LDRD-funded projects. “Searching a design space of this size just wouldn’t be feasible without powerful computing resources because the volume of decisions is too high.”

Experimental validations of top candidates improve countermeasure development

The virtual screening of millions of antibody and antiviral candidates made it possible for the team to identify an initial set of promising compounds in just a few weeks. The next step involved a team of LLNL experts in virology, immunology, and molecular biology, who synthesized and validated the efficacy of the top candidates through experiments—another example of work that was quickly launched by expanding the scope of an existing LDRD-funded project. 

Using binding assays and biolayer interferometry technology, the experimental team conducted precise measurements to test how well the molecules bind to their target. They used these measurements to rank therapeutic candidates and further narrow the pool of computationally identified candidates with the antiviral properties needed to fight this disease.  

The experimental team tailored their research to ongoing computational efforts, so that the two aspects of the research worked synergistically, with validated results feeding back into the predictions.

Multimodal machine-learning tools predict risk factors for COVID-19 patients

Investigators of another LDRD-funded project also shifted their focus after the pandemic began. A team of computational scientists was studying ways to use multimodal machine learning to fuse information from disparate sources and learn the underlying temporal structure of the data. They had already been exploring ways to apply machine-learning tools to existing medical datasets in order to more effectively predict clinical outcomes.

Lead investigator Priyadip Ray realized that his team could adapt this work to conduct predictive risk analyses for COVID-19 patients. Using the dynamic modeling framework they had started developing, the team fused genomic data regarding virus mutations with medical datasets regarding microbes present in clinical samples obtained from COVID-19 patients. 

As their work continues, they are collaborating with data science experts at LLNL to create a diagnostic-testing knowledge base that the research community can use to predict the variables associated with COVID-19 disease progression. The data repository will help medical providers detect unforeseen, potentially actionable changes to a patient’s health and identify optimal treatment options.

Diagnostic tool adapted to screen for COVID-19 co-infections 

LDRD-funded research often focuses on innovative solutions with a broad range of potential applications. One example of this ability to tailor technology initially developed with LDRD funds involves a diagnostic tool known as the Lawrence Livermore microbial-detection array (LLMDA), developed in 2008. This high-throughput, nucleic-acid detection technology can simultaneously analyze and identify up to 12,000 microbial species in a single test. 

LLNL scientists adapted LLMDA to analyze nucleic acids extracted from COVID-19 oral and nasal samples and identify the presence of other pathogens. Their work involved designing a unique COVID-19 signature that could be incorporated into the existing LLMDA technology to detect the presence of the SARS-CoV-2 virus. They validated the new capability using LLMDA to analyze samples that had already tested positive for COVID-19 using more traditional diagnostic methods, and then looked for the presence of other pathogens in the samples. They were able to identify several types of co-infections, such as Influenza A and Streptococcus pneumoniae. 

The research team anticipates that collecting this co-infection data can help scientists better understand how the presence of other pathogens may affect a person’s susceptibility to COVID-19 infection, as well as how they respond when infected with the SARS-CoV-2 virus. 

Related LDRD Projects

• Identifying Potential Antiviral Small Molecules Against the Coronavirus Disease 2019 (PI Felice Lightstone)
• Active Learning for Rapid Design of Vaccines and Antibodies (PI Tom Desautels)
• Rapid Computational Identification of Therapeutic Targets for Pathogens (PI Jonathan Allen)
• Building a Computational and Experimental Rapid Response Pipeline to Counter the Coronavirus Disease 2019 Outbreak and Emerging Biothreats (PI: Brent Segelke)
• A Deep Bayesian Active Learning Framework for Temporal Multimodal Data (PI Priyadip Ray)

Unlocking the mysteries of high-explosive chemistry may hold the key to addressing a range of mission-relevant challenges, including nonproliferation and counterterrorism. During an LDRD-funded research project that recently concluded, LLNL scientists moved beyond computational predictions regarding the behavior of high explosives. They obtained direct measurements of chemical reactions involving an insensitive high explosive—data that is unprecedented in high explosive (HE) science.

The research team conducted the first-ever experiment of this kind, using the world’s most energetic laser to study a high-explosive sample. During their experiments at LLNL’s National Ignition Facility (NIF), investigators captured novel diagnostic data that will help scientists better understand and predict HE behavior. 

According to Lara Leininger, director of LLNL’s Energetic Materials Center (EMC) and principal investigator for this LDRD project, “The laser shot is the first in a series of experiments that will transform the Lab’s understanding of high explosives by producing never-before-captured experimental data quantifying the response of laser-driven high explosives during reaction.”  They captured detailed, temporally and spatially resolved data regarding the evolving chemistry of a reacting insensitive high explosive. 

These expanded diagnostic capabilities will allow LLNL to investigate the structure of HE detonation products during the detonation, while also helping to validate predictive models that play a key role in understanding HE detonation and safety.

During the first phase of the three-year project, which started in 2017, the multidisciplinary team developed novel diagnostic techniques to measure in-situ, dynamic, laser-driven, high-explosive reactions. Their first experiment took place at LLNL’s Jupiter Laser Facility.

According to LLNL physicist Jon Eggert, who led the project’s high-energy-density experiments, the team was able to adapt an existing diagnostic platform, known as TARDIS (TARget Diffraction In Situ), to conduct the experiments. “It was gratifying to see that a platform we developed for very different scientific and programmatic applications, including its dual-probe and large spot-size options, could be used in this new research space,” said Eggert.

In the second phase, they evaluated concepts during a series of shots at the Omega Laser Facility at the University of Rochester. They also investigated target preparation, configuration, and diagnostic setups.

The capstone phase of the project was performed at NIF, integrating the techniques developed in the first two phases with NIF’s advanced diagnostic systems, which enabled the team to extract the required experimental data. 

The sample used in the NIF shot consisted of a non-detonable quantity of less than 7 milligrams of single-crystal TATB (1,3,5-triamino-2,4,6-trinitrobenzene), a HE with low sensitivity to external stimuli such as friction, pressure, temperature, impact, or spark. 

According to Samantha Clarke, lead scientist for the experimental work at NIF, the team used the facility’s long laser drive, coupled with two x-ray probe beams on the same target, to produce clear evidence of the formation of products in less than 50 nanoseconds. The experiment captured a time evolution of products under shock compression exceeding 150 gigapascals (1.5 million times Earth’s atmosphere).

An LDRD-funded team of scientists and engineers developed a new electrochemical reactor paradigm in an effort to fundamentally rethink how chemical reactors are designed, making them smaller, more energy efficient, and more versatile. They also wanted to create an innovative new approach that transforms carbon dioxide (CO2) harvested from the atmosphere into a valuable resource—industrial chemicals, polymers, and other useful compounds.

Much of the chemical-reactor technology in use today involves the use of energy-intensive thermochemical reactors, with an enormous infrastructure and high capital and operating costs. Through several LDRD research projects, including one that wrapped up recently, investigators developed a prototype electrochemical reactor that uses CO2 as the reactant and converts it into multi-carbon molecules, such as ethylene and ethanol. It is small (with an electrode that measures just 1 square centimeter), and it uses a modular platform that can scale up for industrial settings.

The research team integrated computational design, engineering, and materials science to refine the reactor model, along with three-dimensional additive manufacturing techniques, to produce reactor components that offer the microstructure needed to control the catalyst environment. Leveraging the versatility and precise control offered by the additive manufacturing techniques, they can speed up the process to produce electrochemical cells and reactors, and tailor them to specific operational variables. 

LLNL materials engineer Eric Duoss and chemist Sarah Baker led the LDRD-sponsored research team that developed the reactor design and engineered the prototype. Baker is also partnering with Duoss to investigate scale-up of the technology through industry and university collaborations.

Predictive simulation capabilities play a key role in many high-priority missions, including nuclear security, critical infrastructure protection, and biosecurity. The accuracy of these predictive models relies on the ability to compare simulation results with experimental data. However, in many of these mission-critical research areas, it has been extremely challenging to incorporate the full scope of available experimental data into predictive models. 

In an LDRD-funded project, which is part of the Laboratory’s cognitive simulation initiative, researchers are developing advanced machine-learning techniques that will combine our rapidly expanding simulation capabilities with our precision empirical data sets. These learning-based predictive techniques can better align simulations with experimental data, improving the accuracy and efficiency of simulations.

According to Brian Spears, an LLNL physicist who leads the research team, investigators have developed new deep-neural-network technologies designed to make full use of available data sets from experiments. The novel technology is capable of handling extensive collections of multimodal experimental data, including images, vector-valued data, time histories, and scalar measurements. 

These transformative predictive capabilities can speed up extremely complex calculations and compare varied data sources efficiently, without requiring a scientist to scan tremendous amounts of data. The interdisciplinary team that developed this exciting new technology includes experts in machine-learning architectures, deep learning, data harvesting, and intelligent data sampling.

“The ability to combine multiple, scientifically relevant data streams will open the door to a wide range of new types of analyses,” explained Spears. “It will allow us to extract information from our most valuable experimental and simulation data sets that has been inaccessible until now. Fully exploiting this information, in concert with a new suite of related cognitive simulation tools, will lead to improved predictive models.” 

The ability to engineer mission-critical metal parts using advanced manufacturing techniques continues to play a key role in research at LLNL. Existing three-dimensional (3D) printing methods, which build precise parts in sequential layers, do not work well with all types of metals. Recognizing the need to design and deliver metal parts with tailored properties, including parts made from actinide materials, a research team at Livermore adapted a 3D-printing technique known as liquid metal jetting (LMJ) and explored how it can be used to produce parts with the structure and tailored properties needed to meet the needs of national security stakeholders.

According to LLNL physicist Jason Jeffries, who led the LDRD-funded research team, LMJ eliminates problems typically found with selective laser melting (SLM) techniques. “Selective laser melting limits the types of metal feedstocks that can be used because many metals become highly reactive when they are converted to a powder,” explained Jeffries. In addition, the use of fine powders can cause spattering, which may inadvertently contaminate a part, or the powder can get trapped inside a part and vaporize, resulting in an unwanted void in the finished build. 

“Liquid metal jetting is a much gentler and more localized process,” said Jeffries. And because LMJ uses raw feedstock, it can bypass the extra processing step of converting it into powder. 

LLNL’s multidisciplinary effort to enhance LMJ capabilities included researchers in physics, materials science, mechanical engineering, and computational engineering. They explored how to control tiny droplets of metals, heated above their melting temperatures, which subsequently fall onto a substrate to produce 3D parts from a stack of two-dimensional digital patterns. Upon solidification, each layer of jetted droplets acts as a new substrate onto which new droplets are dispensed to create the adjacent layer. As the droplet-on-demand process repeats, complex 3D geometries can be printed with extremely fine detail. 

The team studied how various techniques affect the build speed and quality of parts. For example, they developed and tested a machine with two different printing modes—a continuous stream and a droplet-on-demand process, which produces discrete droplets ejected from a nozzle. By combining the capabilities into a single machine, an object can be manufactured using both modes, offering the speed of a continuous material stream, and the resolution offered by the droplet-on-demand mode.

The team also studied how to modify the process to obtain optimal mechanical properties. For example, they analyzed how much of the underlying structure should melt as the molten drop bonds to the metal underneath it. In addition, they explored ways to control the cooling rate of the microstructure by modifying the droplet ejection velocity, size, and frequency.

Researchers demonstrated that the technique can be used to produce metal parts with unique shapes, structures, and properties that are not possible with conventional machining, casting, or 3D-printing methods. 

The summer of 1945 marked a turning point in history, with the world’s first nuclear detonations. When editors of a U.S. Army journal decided to publish a special edition commemorating the 75^th anniversary of these historic events, they contacted LLNL nuclear forensic scientist Kim Knight, seeking possible articles for a special issue of the Countering Weapons of Mass Destruction (WMD) Journal. 

Knight has been studying historic fallout from near-surface nuclear events for many years, and at the time the editors contacted her, she was already working with more than 25 LLNL investigators on an LDRD-funded research initiative studying how different environments affect nuclear explosions. 

When the special edition of the journal was released in late 2020, 5 of its 15 articles were written by LLNL scientists.

“The timing of our work fit perfectly for their request,” said Knight. “It was a great way to bring awareness of our work to a broad audience, including our hypothesis that interaction of the local environment with a nuclear detonation changes how the fireball evolves.”

Knight’s research team combined computational and experimental approaches to create an informed framework capable of modeling nuclear detonations, taking into account how the environment contributes to the physical and chemical evolution of the fireball plasma and resultant debris. 

Investigators analyzed historical films of nuclear tests and studied nuclear debris; modelers built into simulations of nuclear detonations the key physics and chemistry needed to account for time-dependent entrainment; and experimentalists studied chemical speciation and early debris-formation processes through laser interactions with matter and plasma chemistry.

“The experimental work informs the chemistry that we need for the models, which we’re grounding in historical data from films and debris,” Knight explained.