Mengjun Shu: Mapping tree resilience with AI, biological data

Deep inside a tree's genes, its chemistry and its hidden partnerships with microbes lie the secrets to survival. Mengjun Shu is uncovering the hidden biology of tree resilience, using big data and computational methods such as statistical modeling and artificial intelligence to build better feedstocks for factories producing advanced chemicals and materials.

"Trees are stuck in their environment no matter the conditions - they can't migrate as some other species do," said Shu, a staff scientist in the Biosciences Division at the Department of Energy's Oak Ridge National Laboratory. "To help us figure out survival strategies for trees in different growing conditions, I make connections between biology and data." Her work is aimed at the development of biomass crops such as hardy, fast-growing poplar trees, advancing biotechnology for U.S. manufacturing and agricultural sector success.

Shu's modeling and AI work relies on large datasets from sources such as genome-wide association studies, which link tree traits to genetic variants, as well as studies of gene activity and data on small molecules called metabolites that can serve as early stress indicators and play a role in trees' natural defenses. She's also using insights gleaned from hyperspectral images of plants as they grow and move through ORNL's automated Advanced Plant Phenotyping Laboratory (APPL), revealing differences in plant chemistry and structure beyond what the eye can see.

One project she's leading explores chlorophyll-containing green woody tissue in poplar trees, and how that tissue may relate to drought response and seasonal adaptation, part of her work for the DOE Center for Bioenergy Innovation at ORNL. "People often focus on leaves when they're studying plant photosynthesis," Shu said. "But woody tissues can also contain chlorophyll [the pigment that absorbs energy from sunlight], and contribute to stress-response biology in plants."

Using data and AI to detect biological mechanisms at work

In her research for the DOE Plant-Microbe Interfaces Science Focus Area, Shu's goal is to understand how poplar genetic variation shapes microbial communities and plant performance under stress, especially drought. She uses statistics to reveal how genetic variation influences plant traits and how patterns of gene activity can help scientists predict plant response to beneficial microbes.

Shu's use of AI is accelerating discoveries in several of her projects. "My AI-related work is mostly focused on how genome-scale models, including DNA language models, can detect biological signals across plant species and eventually connect genome sequence with photosynthetic function and environmental response," Shu said. "If we can identify meaningful patterns in data, we can generate better hypotheses about the biological mechanisms at work."

Shu is mining the treasure trove of data gathered from hyperspectral and other advanced imaging in the high-throughput APPL facility to better observe plants as they grow. APPL's robotic platform gathers more data and analyzes it faster using AI than scientists taking measurements by hand and feeding data into computers.

She used APPL's automated phenotyping to accurately predict metabolite profiles in poplar under drought stress, as outlined in an Environmental and Experimental Botany paper last fall.

"Plants may look normal to our unaided eye, but they may already be stressed. The hyperspectral data in APPL catches what our eyes can't in terms of water and other stress plants may encounter," Shu said.

In another project, she used statistical analysis to compare millions of genetic differences across hundreds of poplar trees, identifying DNA variants consistently associated with wood composition traits. She and colleagues used computation to then narrow hundreds of genetic signals to three strong candidate genes involved in wood formation, as described in New Phytologist.

Creating high-quality phenotyping data for better biological connections

Shu is interested in making high-throughput plant data easier to understand. APPL can follow hundreds of plants through drought and recovery, producing thousands of measurements from images and sensors. But a change in a sensor signal is most useful when scientists know what it means inside the plant. Shu wants to connect those signals with direct measurements of water use, photosynthesis and metabolites, helping researchers tell whether a plant is conserving water, changing its chemistry or recovering from stress. That biological context can make it easier to compare results across experiments, including greenhouse and field studies.

"With this approach, we can address a major bottleneck in modern plant science: converting high-throughput measurements like hyperspectral imaging into biologically interpretable insights," Shu said. "We have come so far with plant genetics that it is no longer a main limitation in understanding and developing better plants. The bigger challenge as I see it is high-quality, interpretable phenotype data, especially in the field."

She describes her long-term vision as connecting genomic patterns to photosynthetic performance and environmental adaptation, eventually enabling direct answers to questions such as: "Given this tree's genome, in what environment would it best perform?"

Shu emphasized how ORNL's multidisciplinary environment supports her work. "The big questions we're trying to answer require many types of expertise to come together, including genetics, metabolomics, high-throughput phenotyping, field trials, microbial ecology and validation work," Shu said. "ORNL is a strong place for this research because that team science environment already exists here."

Her advice for young scientists?

"Collaboration is key," Shu said. "If you have an idea, talk to your colleagues and get their perspectives and insight. For high-quality projects and to publish in high-impact papers, you need to tap into different disciplines and expertise."

She also noted that it's getting easier to learn technical skills, including the use of AI to help debug models. But here she adds a note of caution. "People still need to spend time learning biology. No matter how fancy the model is, if we don't understand biology, we can't interpret the results."

UT-Battelle manages ORNL for DOE's Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science. - Stephanie Seay