Scientists at the U.S. Department of Energy’s (DOE) Argonne National Laboratory were recognized for their achievements in high performance computing (HPC) with three HPCwire Awards. The awards were announced at the SC23 conference.

Best Use of HPC In Life Sciences

An Argonne-led team won the Readers’ Choice Award for Best Use of HPC in Life Sciences for the development of APACE, a computational-framework-as-a-service that optimizes AlphaFold2 to run at scale on the Delta supercomputer. This machine is housed on the University of Illinois Urbana-Champaign (UIUC) campus at the National Center for Supercomputing Applications (NCSA). APACE was so successful that it reduced time-to-solution for off-the-shelf AlphaFold2 predictions from days to minutes.

The team for this award includes Hyun Park (Argonne & Beckman/UIUC), Parth Patel (Argonne & UIUC), Roland Hass (NCSA/UIUC) and Eliu Huerta (Argonne & University of Chicago).

AlphaFold2 is an artificial intelligence (AI) model that accurately predicts protein structures, though it was not designed to run optimally in HPC environments. Eliu Huerta — Argonne Lead for Translational AI and team lead on this project — explained the true significance of this work in an interview after the awards.

“The original AlphaFold2 model is computationally demanding, requiring significant computational resources and time for accurate predictions,” Huerta said.

Huerta explained that the process faced limitations in real-time protein structure prediction due to the need for CPU-based tasks like multiple sequence alignment and template search, while GPUs were used for structure prediction. This makes it less suitable for time-sensitive applications or hardware with limited computational power.

To solve this problem, the APACE framework was designed from the ground up to run optimally and at scale in modern HPC platforms. The team optimized CPU and GPU performance, leveraged solid state drive data storage and used DDN’s Infinite Memory Engine data staging to effectively handle AlphaFold2’s TB-size database.

Best HPC Collaboration

The Readers’ and Editors’ Choice Awards for Best HPC Collaboration were awarded to researchers from Purdue University, Argonne and Rolls-Royce LibertyWorks. This group leveraged high-fidelity computational fluid dynamics (CFD) modeling and HPC to inform design strategies for integrating advanced hydrogen-fueled rotating detonation combustors within stationary gas turbines.

Specifically, the team — led by Professor Guillermo Paniagua from the School of Mechanical Engineering at Purdue — worked on the design and testing of a hydrogen-air rotating detonation combustor (RDC) with a new class of diffusive nozzle guide vanes to enable coupling with a Rolls-Royce M250 engine for power generation. These nozzle guide vanes direct the airflow onto the turbine blades of the engine while also converting pressure energy into kinetic energy.

Pinaki Pal, a senior research scientist at Argonne, explained the work further:

“We leveraged the Bebop supercomputer at Argonne’s Laboratory Computing Resource Center to perform high-fidelity 3D CFD simulations, providing insights into combustion physics within the Purdue hydrogen-air RDC,” Pal said. Purdue University, on the other hand, utilized lower-order CFD simulations performed on their Rosen Center for Advanced Computing’s Bell supercomputer to optimize the design of the combustor and turbine components to be integrated with the RDC. Rolls-Royce LibertyWorks offered valuable guidance and also provided an M250 helicopter engine and testbed to Purdue for concept demonstration of this integrated RDC-turbine design.

According to Pal, ​“The implications of this research are far-reaching and transformative. This work will provide unique guidelines and best practices to the industry for efficient transition of high-speed, unsteady flow from RDC exit into a turbine rotor for reliable work extraction. This will enable the integration of advanced hydrogen-based pressure gain combustion technology into stationary gas turbine plants for decarbonized power generation.”

Best Use of High-Performance Data Analysis & AI

Argonne earned the Editors’ Choice Award for Best Use of High-Performance Data Analysis & Artificial Intelligence in collaboration with the University of Chicago, the University of Illinois at Urbana-Champaign, and the University of Illinois Chicago. The team includes Hyun Park (Argonne and Beckman/UIUC), Xiaoli Yan (Argonne & UIC), Ruijie Zhu (Argonne & UChicago), Eliu Huerta (Argonne & UChicago), Santanu Chaudhuri (UIC & Argonne), Ian Foster (Argonne & UChicago), and Emad Tajkhorshid (UIUC).

Huerta was again team lead for this research, and spoke about how the team combined generative AI, high-throughput screening methods, and large-scale molecular dynamics simulations to model novel, stable and high-capacity metal-organic frameworks (MOFs) for carbon capture within seconds.

Traditional methods to discovering MOFs include experimental approaches and the use of molecular dynamics simulations, with differing levels of detail and accuracy, Huerta said.

“There has been a lot of work harnessing different flavors of AI to design MOFs,” Huerta said. ​“The value of our work is that we have combined different fields of research to provide a fresh approach to the problem. To be specific, we combined generative AI, materials science, drug design and discovery, high throughput screening, molecular dynamics simulations and Grand Canonical Monte Carlo simulations to create and then validate our methods.”

This computational framework can produce 120,000 AI-generated MOFs in about 30 minutes using a single node in the Theta supercomputer at the Argonne Leadership Computing Facility (ALCF), a DOE Office of Science user facility. Then, it leverages distributed computing to screen and validate the properties of these MOFs. In about 12 hours, using the ALCF’s Polaris supercomputer, this AI-coupled-HPC workflow can produce a novel set of AI-generated MOFs with optimal performance for carbon capture.

“Compare this to a single Grand Canonical Monte Carlo simulation, which we also used to quantify the capacity of a single MOF to capture carbon dioxide, and which is completed within six hours using Polaris,” said Huerta. ​“In brief, we are reducing time-to-insight dramatically, and demonstrating that knowledge in different fields may be cross pollinated to provide entire novel solutions to timely and relevant energy grand challenges.”

“These advances in AI are only possible by the ingenuity of bold and visionary researchers, and on that note, I want to highlight the work done by Hyun Park, Parth Patel, Xiaoli Yan, and Ruijie Zhu. They are brilliant students whose innovative work at the interface of AI and supercomputing is disrupting how science is done,” Huerta said.