SI Pilot Project: Pacific Northwest National Laboratory
- Principal Investigator: Jaydeep Bardhan (Pacific Northwest National Laboratory; PNNL)
- Co-Investigators: Aivett Bilbao (PNNL, Environmental Molecular Sciences Laboratory; EMSL), Song Feng (PNNL), Jeff Czajka (PNNL), Joonhoon Kim (PNNL), Priscila Lalli (PNNL), Vivian Lin (PNNL), Lummy Monteiro (PNNL), Kyle Pomraning (PNNL), Ravi Ravichandran (PNNL, EMSL) Natalie Sadler (PNNL), subcontract to Kjiersten Fagnan (Lawrence Berkeley National Laboratory, Joint Genome Institute)
- Scope/Objectives: Develop super intelligence (SI) agent technologies and autonomous workflows to leverage DOE-BRIDGE, integrating automated high-throughput experimental data.
- Potential Impact and Interface with the American Science Cloud (AmSC) and Transformational AI Models Consortium (ModCon): Develop SI models as Model Team and connect with AmSC through BRIDGE.
- BER Program Manager: Ramana Madupu
Automated experimental platforms offer enormous opportunities for generating phenotypic data at scale for training predictive super intelligence (SI) models, especially when integrated with BER’s extensive genomic data resources.
This project aims to develop emerging SI agent technologies and autonomous workflows to leverage DOE-BRIDGE in integrating automated high-content, deep phenotyping experimental capabilities, ultimately creating a unified framework that can effectively interpret complex biological responses when used at scale with automated laboratory platforms. The team will use the new Anaerobic Microbial Phenotyping Platform (AMP2) system at the Environmental Molecular Sciences Laboratory as the use case automated platform and its mass spectrometry (MS) and cell imaging data as two exemplars of phenotypic data.
Work centers on the soil microbe Pseudomonas putida due to its biotechnological potential and relevance as a model system for critical minerals and materials research. For both MS and cell imaging data types, knowledge graphs and SI agents form the critical enabling technologies for data integration and ultimately experiment design, leveraging genomic data and genome-scale metabolic models.
The project’s short-term goal is to integrate deep phenotyping data into cohesive predictive models, leveraging BRIDGE data resources to support autonomously analyzing multiomics data, developing metabolic models, and designing experiments, accelerating the optimization of critical minerals and materials extraction processes.
Project Structure
One major project thrust focuses on MS-based proteomics and metabolomics. The team has built an initial prototype of a knowledge graph and agentic system that can be used for experiment and design and is now focused on expanding the knowledge graph with relevant MS embeddings.
A second major thrust focuses on cell imaging, specifically the imaging capabilities in AMP2’s flow cytometer. Paralleling the concept in the MS thrust, the team aims to take phenotypic signatures from SI vision models (whether fine-tuned foundation models or otherwise) and integrate them into knowledge graphs that SI agents can use for experiment design.
These two activities are supported by a third focus on key infrastructure enabling integration: data models, metadata schema, and pipelines for interaction with BRIDGE. The project has a subcontract with the DOE Joint Genome Institute to support development of needed infrastructure.
Overall, this work will advance the use of SI for autonomous science on state-of-the-art automated experimental platforms, enabling integration of BER’s genomic data resources with new phenomic data types (MS and imaging data) in BER’s federated data resources to optimize experimental design-build-test-learn cycles.