The Transformational AI Models Consortium (ModCon) is fostering the development of transformative AI capabilities through initial projects being conducted by collaborative teams across the DOE national laboratory complex. These seed projects are driving advancements in AI across a diverse range of critical areas. PNNL is proud to play a key role, contributing to six of the initial projects. 

Foundational AI Models for Optimizing and Understanding Biological Systems (FAMOUS)

This project will transform biology into the ultimate design and manufacturing engine by leveraging advanced AI to decipher protein and pathway function, automating data generation and collection, and integrating DOE’s experimental and supercomputing systems with biotech and AI companies. The team will enable synthetic biology approaches that could reinvent manufacturing, agriculture, and resource management. 

PNNL will build and fine-tune metagenome DNA and mass spectrometry language models to dramatically accelerate time-to-engineering for biosystems. These models will be cross trained with AI models developed with our FAMOUS partners, resulting in multimodal foundation models that “translate” science inquiries across the languages of mass spectrometry peaks, gene and protein sequences, and functional characterization. Ultimately, agentic forms of these models will be used to accelerate biological discovery and engineering as the “central nervous system” of advanced automated profiling and discovery laboratory platforms.

PNNL POC: Chris Oehmen 

AI for Planning and Operation of the U.S. Power Grid

This project seeks to infuse AI foundation models and agentic systems into power system planning and operations to accelerate the secure and reliable modernization of the U.S. power grid. The team will prepare and integrate over 200 TB of multimodal grid data and release a large-scale, physics-informed foundation model. 

PNNL leverages its leading utility and grid modeling capabilities to integrate physics-based simulations with AI-driven learning frameworks, developing scalable architectures to train and validate foundation models and agentic applications using diverse multimodal power system data. The lab also advances AI assurance, cybersecurity, and model interpretability to ensure trustworthy deployment in real-world grid operations. 

PNNL POCs: Marcelo Elizondo and Xiaoli Duan

CM²US: Critical Minerals and Materials to Unlock Supply

This project endeavors to secure the domestic supply chain of critical minerals and materials by revolutionizing the discovery, development, and production of these vital resources as one connected system. The team will create an AI-driven engine that acts as a strategic guide for the entire supply chain, producing a dynamic digital map of the entire process from mining to market.

PNNL will extend the Global Change Analysis Model, developed in-house, to represent full CMM supply chain dynamics from extraction through manufacturing, providing the benchmark for comparing graph neural networks, large language models, and diffusion models developed across the other labs. 

PNNL POC: Nancy Washton

AI-Driven Co-Design for Microelectronics

This project seeks to transform the way microelectronics research is conducted to unlock new pathways for materials research and computer systems design. The team will significantly reduce the time to develop accelerators as needed by DOE science applications.

PNNL is developing a generative AI agentic workflow that leverages new MLIR compiler features to interface to our Software Defined Architectures (SODA) hardware generation capabilities. Our goal is to support domain scientist development of customized edge computing hardware designs that support their needs for measurement and observation data.

PNNL POCs: Jim Ang and Robert Rallo

AI for Efficient Quantum Algorithms 

This project will accelerate the discovery of new quantum algorithms that move beyond quadratic speedups and open the door to exponential or domain-transformative advantages. The team will create an integrated suite of AI-enabled tools and methodologies that advance the frontiers of quantum computer science and speed the discovery of quantum algorithms.

PNNL will work with other lab partners to develop novel quantum algorithms and verification systems for searching and other operations of quantum data by leveraging large language models and other advanced generative AI models through the center. 

PNNL POC: Ang Li

Accelerating Scientific Discovery through AI-Driven Code Development 

This project aims to port and optimize complex flagship DOE applications using AI to suggest algorithmic improvements, generate efficient code for GPUs and accelerators, and validate performance by creating an AI-driven co-development ecosystem to shorten the time from scientific concept to validated simulation.

PNNL seeks to develop an AI agentic system to accelerate scientific discovery through AI-driven code development of computational chemistry methods. By learning from decades of DOE codes, compiler traces, and performance data of quantum chemistry frameworks, such as NWChem, AI systems will create a co-development ecosystem to generate, optimize, and verify high-performance and energy efficient-scientific software. The team will focus on advanced reasoning and accuracy of the models, as well as their scalability and energy efficiency.

PNNL POC: Roberto Gioiosa