AI Predicts DNA Binding With Hypercomplexity, NC State Finds

North Carolina State University researchers have developed an artificial intelligence model capable of predicting which DNA molecules bind to one another with 83.5% accuracy, a significant improvement in understanding the intricate relationships within genetic systems. The new model, named BINND: Binding and Interaction Neural Network for DNA, was trained on a dataset of 144 million sequence pairs, allowing it to capture a level of complexity previously unattainable with biophysical modeling. “We often think about binding as a simple relationship – Molecule A binds to Molecule B,” says Albert Keung, co-corresponding author of the study and the Goodnight Distinguished Scholar in Innovation in Biotechnology and Biomolecular Engineering, “But in biological systems, it’s far from simple.” This advance, according to the team, has potential applications ranging from more sensitive biomedical diagnostic tools to novel DNA computing systems.

Deep Learning Model BINND Predicts DNA-DNA Binding Accuracy

Achieving 83.5% accuracy in predicting DNA pair binding, BINND represents a substantial improvement over previous predictive tools. These tools struggled with the inherent complexity of DNA interactions, often relying on smaller datasets and biophysical modeling that failed to capture the full scope of binding relationships. “We took a different experimental approach that allowed us to generate substantially more data on which DNA sequences bind to each other,” explains Karishma Matange, a Ph.D. graduate of NC State and co-lead author of the paper published in Nature Communications. This expanded dataset enabled the team to move beyond extrapolation and leverage the pattern-recognition capabilities of deep learning. The model’s tendency to err on the side of predicting no binding, rather than falsely predicting binding when none exists, is a notable characteristic of its performance. “BINND is at least 10% more accurate than the best existing model,” states Gunavaran Brihadiswaran, co-lead author and Ph.

Beyond improved accuracy, the researchers utilized BINND to create a database illustrating the hyperconnected nature of DNA-DNA binding, mapping interactions between 96 distinct 20-character DNA sequences and 26 others. Molecule A may bind to dozens of other molecules, to varying degrees, Keung further notes that this database holds particular promise for advancing DNA computing, providing critical information for capturing and retrieving data using DNA.

144 Million Sequence Pair Dataset Enables AI Training

The pursuit of understanding DNA interactions has entered a new phase, moving beyond limited datasets and increasingly relying on artificial intelligence to decipher the complex relationships between sequences. Previous predictive models struggled with the inherent complexity in biological systems, often extrapolating from small datasets and biophysical modeling; however, researchers are now leveraging a significantly expanded resource to train more accurate AI. A team at North Carolina State University has created a database comprising 144 million sequence pairs, a substantial leap forward in the volume of data available for analysis. The shift toward AI-driven prediction is driven by the realization that DNA binding isn’t a simple one-to-one correspondence. Achieving 83.5% accuracy, and notably, when errors occurred, the model tended to predict no binding rather than a false positive.

This enhanced predictive capability has implications extending beyond fundamental biological understanding, potentially impacting fields like biomedical diagnostics and DNA computing. “Capturing that complexity is also critical if we want to develop tools that make full use of biomolecules, such as diagnostic tools that are sensitive to genetic differences or DNA computing systems that rely on DNA to store and retrieve data,” Keung states, highlighting the broad applicability of this research. The team has made BINND publicly available on GitHub, hoping to foster further innovation within the scientific community.