Pacific Northwest National Laboratory (PNNL) scientists are highlighting the potential of artificial intelligence and exascale computing to revolutionize biological research, based on findings from the 2025 Workshop on Envisioning Frontiers in AI and Computing for Biological Research. Researchers detailed how these advanced technologies can enhance biological studies, identifying four priority research directions including multimodal data assembly and AI-enabled experimental systems. Kirsten Hofmockel, who leads the Soil Microbiome Science Focus Area project at PNNL, notes a key challenge: “One of the big challenges I routinely run into collaborating across domains of science is integrating diverse data across multiple scales to establish genotype to phenotype relationships.” This report underscores the importance of combining biological data collection with leading computational capabilities to accelerate scientific discovery.
Workshop Highlights AI for Multiscale Biological Research
Participants focused on applying techniques like multiscale modeling to accelerate discovery. A central theme emerging from the workshop was the challenge of connecting genotype to phenotype, a task complicated by the sheer diversity and volume of biological data. She further emphasized that data arrives “in various formats, from images to genetic sequences,” necessitating meaningful integration for effective AI applications. Neeraj Kumar, Chief Data Scientist at PNNL, highlighted the transformative potential of this interdisciplinary approach, stating, “AI and advanced computing hold immense promise to unlocking breakthroughs in biological research.” He envisions a future where “close collaboration between computer scientists and domain scientists” leads to co-designed systems for next-generation scientific discovery, driving innovation in biotechnology and biomanufacturing.
Integrating BER Data with ASCR Computational Capabilities
The convergence of biological data generation and advanced computing is rapidly reshaping scientific inquiry, with the Department of Energy’s Biological and Environmental Research (BER) program and Advanced Scientific Computing Research (ASCR) program at the forefront. A recent workshop detailed how combining BER’s data collection with ASCR’s computational power—including exascale architectures—represents “an important path to progress.” This synergy isn’t merely about processing power; it’s about enabling entirely new research avenues previously constrained by data integration challenges. Researchers identified four key areas for advancement: multimodal data assembly, multiscale biosystems simulation, AI-enabled experimental systems, and novel algorithms for genomics. Both Kumar and Hofmockel concur that progress hinges on algorithmic innovation and collaborative efforts incorporating biology, computing, and automation to unlock biological mechanisms and support biotechnology.
Genotype-Phenotype Relationships Drive Algorithm Innovation
Neeraj Kumar, chief data scientist at Pacific Northwest National Laboratory (PNNL), is spearheading efforts to unify infrastructure for autonomous discovery across biology, chemistry, and critical materials. He’s integrating the Transformational AI Models Consortium with the American Science Cloud, aiming to bridge computational and domain sciences for national impact. Both Kumar and Hofmockel agree that innovation in algorithms and AI is crucial for interpreting diverse biological data, with Hofmockel stating, “We need to innovate algorithms and leverage AI to integrate and interpret diverse biological data.” New collaborations incorporating biology, advanced computing, and automation are seen as vital for advancing biotechnology and biomanufacturing.
The integration of multimodal data necessitates sophisticated deep learning architectures, moving beyond simple concatenation of data streams. Technologically, this involves mapping disparate biological inputs—such as single-cell RNA sequencing data, electron microscopy images, and genomic fingerprints—into shared, low-dimensional embedding spaces. This process allows AI models to discover latent connections between features that exist in different formats, facilitating predictive modeling where correlations might otherwise remain siloed within their original data modality.
Exascale computing provides the necessary computational substrate for complex molecular dynamics simulations that are infeasible on current petascale systems. Specifically, simulating protein folding or metabolic pathways at the quantum mechanical level requires massive parallelization across thousands of specialized processing cores. These simulations enable researchers to predict functional changes of novel biomaterials under various environmental stresses, drastically reducing the need for iterative, costly wet-lab physical experimentation.
The concept of AI-enabled experimental systems points toward fully closed-loop automated research environments. Here, machine learning algorithms operate as the central intelligence, interpreting initial data and then automatically designing and executing the next set of physical experiments. The system uses robotic platforms, imaging hardware, and microfluidics to rapidly test hypotheses, accelerating the iterative design-build-test cycle and maximizing data throughput far beyond human capacity.
To effectively model the genotype-phenotype relationship, researchers are increasingly turning to advanced graph neural networks (GNNs) and causal inference models. Unlike standard deep learning, which excels at correlation, GNNs model the complex, non-linear interactions within biological networks, treating genes, proteins, and metabolites as interconnected nodes. This shift allows for the development of predictive regulatory maps, moving the focus from merely observing association to computationally establishing biological causality.
“We need to innovate algorithms and leverage AI to integrate and interpret diverse biological data.”
