Clemson researchers will use AI to explore how plant cells respond to stress - news.clemson.edu

When crops face drought, extreme heat or disease, every part of the plant does not respond in the same way.

Through a three-year, $830,055 grant from the National Science Foundation, Clemson University researchers will use artificial intelligence to study how individual plant cells control their genes.

The answers could eventually help researchers and breeders develop more resilient crops.

“Different cells may function differently in different environmental conditions, developmental stages or under stress,” said Shahid Mukhtar, a professor in the Department of Genetics and Biochemistry who is leading the project. “If we’re going to figure out precision genetic editing, we need to first learn gene regulation at the single-cell level.”

And being able to advance precision genetic editing “could eventually help researchers target the cells most important to drought tolerance, heat resilience or disease defense,” he explained.

Different genes

Gene regulation is the system cells use to decide which genes to use. A cell in a leaf may need a different set of active genes than a cell in a root. A plant under stress during a drought may activate different genes than a well-watered plant. A cell responding to a pathogen may behave differently than a nearby cell that has not detected the danger.

For decades, scientists studied gene activity by studying entire tissues. Researchers would take a whole leaf or root, grind it up and measure the combined activity of all its cells. This “bulk RNA” approach can reveal broad patterns, but it can also hide important differences among cell types. This approach also cannot easily show which specific cells are responsible for a particular response.

New technologies allow researchers to study genetic activity in individual cells.

Molecular switches

Transcription factors are proteins that act like molecular switches. Plants have a couple thousand transcription factors or regulators, which bind to specific regions of DNA and can increase or decrease a gene’s activity, influencing how a cell develops or responds to a challenge. The Clemson project will use AI to infer those relationships and generate hypotheses that can be tested experimentally. Some of the approaches they plan to test have already been used in animal research but have not been widely applied to plants.

The first phase of the project will evaluate those existing AI methods, particularly tools known as interpretable or explainable AI. Unlike a computer system that simply produces a prediction, explainable AI is designed to help researchers understand the reasoning behind its output.In the second phase, Deepanshu Verma of the Clemson School of Mathematical and Statistical Sciences will develop new mathematical models to capture how these gene regulatory relationships change over time.

The first approach, flow matching, treats gene expression as a continuous process governed by differential equations rather than a series of frozen snapshots, making it possible to identify which molecular switches are actively driving a change rather than simply accompanying it.

The second approach, generative flow networks, considers thousands of possible regulatory network structures simultaneously and assigns each a measure of confidence based on biological evidence. This shifts the analysis from a single best-guess network to a ranked set of predictions, allowing researchers to focus experimental resources on the connections most likely to hold up in the lab.

“Most existing tools give you one answer,” Verma said. “What we are building gives you an answer and tells you how much to trust it, which is what you actually need before you go into the lab.”

Model plant

The researchers will begin with Arabidopsis, a tiny flowering weed that is widely used in plant biology labs because it grows quickly, takes up little space, produces many seeds and has a life cycle of about six weeks.

The third phase of the project will use computer-based simulations to predict what could happen if a major transcription factor is removed. The researchers will then test those predictions with real genetically modified plants under different stress conditions.

The research could eventually support a more precise form of crop improvement, Mukhtar said. If scientists can identify genetic switches that work only in certain types of cells, they may be able to develop targeted changes for drought tolerance, heat resilience or disease defense by focusing on the cells most important to specific traits.

Mukhtar also plans to create an Artificial Intelligence and Genomics Teaching Camp (AGTC). The acronym AGTC reflects the four chemical bases that make up DNA: adenine, guanine, thymine and cytosine.

The camp will introduce high school students to AI and genomics, fields that are becoming increasingly connected. Students will have the opportunity to learn how biological data can be used to understand living systems and how computing can help scientists make sense of information too large and complex to study by hand.

“AGTC will give students an opportunity to see how biology and computing can work together to answer important questions about the living world,” Mukhtar said. “By introducing students to AI and genomics early, we hope to encourage their curiosity and help prepare the next generation of scientists.”