Sandia National Labs FY20 LDRD Annual Report

FY20 ANNUAL REPORT

Diversified therapeutic phage cocktails from close relatives of the target bacterium. Bacteriophages are viruses that evolved to attack and kill specific bacteria, making them promising candidates for treating bacterial infections. This approach, known as phage therapy, is more urgent now that bacteria are developing resistance to antibiotics. Software tools developed in this project by Sandia, the University of Florida, and New Mexico Tech collaborators, helped discover new bacteriophages for pathogenic bacteria by searching the genomes of these bacteria for viral genetic sequences that the bacteria have incorporated into their own. This is a significant improvement in our ability to identify bacteriophages, which used to require searching through environmental samples collected in the field. Upon identification, viruses are reconstructed in the laboratory from their genetic blueprints and modified to enhance their lethality to the target bacteria. Choosing

bacteriophages from close relatives of the target makes them more effective. Testing against infections in model systems has demonstrated the potential of this approach. With further testing, these results promise quick development of countermeasures that specifically target any bacteria emerging as national security biothreats. (PI: Kelly Williams)

Method for countering any bacterial pathogen using bacteriophages discovered in their genomes.

Energy-efficient implementation of partial differential equations by stochastic and deterministic neuromorphic algorithms. Neuromorphic computing is a new approach to achieving low-power computing solutions by leveraging inspiration from the brain. Since it is critical to stockpile stewardship to facilitate low-power neuromorphic architectures, the team researched computational foundations for solving partial differential equations (PDE). They developed neural algorithm approaches for Monte Carlo-dependent mission relevant PDEs that

can be mapped to emerging brain-inspired computer architectures and demonstrated a random walk algorithm with an overall energy consumption (speed/power) that is 20X to 100X better than conventional computing. This is the first formal demonstration of a neuromorphic advantage on a numerical computing application. (PI: James Bradley Aimone) While estimating a solution to a 2D diffusion equation, this figure represents the simulation of 1,000 random walkers moving in parallel over a torus-shaped mesh that represents the networked connections on the Intel Loihi system. The entire random walk process is performed exclusively on the neuromorphic platform.

46

LABORATORY DIRECTED RESEARCH & DEVELOPMENT