When Gunavaran Brihadiswaran started his Ph.D. at North Carolina State University, he faced a puzzle that had stumped scientists for years: predicting how DNA molecules stick to each other. DNA doesn't work like a simple lock and key. A single DNA sequence might bind to dozens of others, in different ways, under different conditions. Capturing that messy reality matters deeply — for building better medical tests that detect genetic diseases, and for an emerging field called DNA computing, where researchers hope to store enormous amounts of data inside biological molecules.
"Molecule A may bind to dozens of other molecules, to varying degrees," says Albert Keung, an associate professor of chemical and biomolecular engineering at NC State who co-led the research. "Capturing that hypercomplexity is a significant challenge, but it is critical if we want to better understand natural genetic systems."
The team took a different approach than previous efforts. Rather than relying on small datasets and mathematical predictions, they built a massive library of 144 million DNA sequence pairs and their binding behaviors — by far the largest dataset of its kind. They used this to train an artificial intelligence model called BINND, short for Binding and Interaction Neural Network for DNA. The AI learned to spot patterns across millions of examples, much like how a speech recognition program learns to understand accents after hearing thousands of recordings.
In testing, BINND predicted which DNA sequences would bind together with 83.5% accuracy — and when it did make mistakes, it usually erring on the side of saying two sequences would not bind when they actually did. The researchers say BINND outperforms the previous best tool by at least 10%, a meaningful gap in a field where even small improvements matter.
To show what the model can do, the team used BINND to create a database mapping how 96 different 20-character DNA sequences interact with 26 others. This kind of detailed map could help engineers design DNA systems that reliably store and retrieve information — a major hurdle for scaling up DNA data storage.
"We're hoping that others in the research community will make use of BINND," says James Tuck, a professor of electrical and computer engineering at NC State who co-authored the study. The team has released BINND freely on GitHub, an online platform where scientists can download and build on the tool.
The researchers acknowledge that 83.5% accuracy is not perfect — DNA binding can depend on factors beyond just the genetic sequence itself. But they say the work represents real progress, opening a window into a biological complexity that scientists have long struggled to see clearly.