EAGER: NAIRR Pilot: IMPRESS: Integrated Machine-learning for PRotein Structures at Scale

The Integrated Machine-learning for PRotein Structures at Scale (IMPRESS) project aims to harness the power of artificial intelligence (AI) and high-performance computing (HPC) to revolutionize the way we design and validate proteins tailored for specific purposes. Creating novel proteins can potentially transform numerous aspects of human life. IMPRESS will address fundamental challenges in AI-driven protein design, including determining optimal neural architectures, efficient training of foundational models, and integrating diverse data sources such as experimental and simulation data. This will enhance the accuracy and efficiency of protein design and provide the necessary computing capabilities to pave the way for future research and development. The project will provide valuable training opportunities for students and early-career researchers. By enabling the effective creation of high-quality tailored proteins, the project can potentially deliver many tangible benefits to society.Artificial intelligence and computing advances have set the stage for designing novel proteins tailored for specific purposes. However, the space of possible protein sequences and structures is astronomically large, even for modestly long proteins; thus, obtaining high convergence between generated and predicted structures requires significant computational resources in sampling. Coupling AI systems with traditional HPC simulations promises significant scientific acceleration, defined as the number of high-quality structures for a given computational cost. The Integrated Machine Learning for Protein Structures at Scale (IMPRESS) project will enhance our ability to tailor proteins by designing and implementing advanced systems that support the online coupling of AI with HPC tasks. Specifically, this project will accelerate the evaluation of possible protein sequences over “vanilla” approaches that do not leverage the online coupling of AI and HPC capabilities. The integrated AI-HPC infrastructure and methodology will provide the ability to “evaluate as you go” the effectiveness of models and evolve the specific set of simulations used to generate data and train models. IMPRESS will also enable novel modes and methods in online coupling and concurrent execution of AI and HPC on the NAIRR platform.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.