Integrating Sequence, Structure, and Dynamics for Predicting and Designing Nucleic Acids

The central dogma in biology describes the flow of genetic information from DNA to RNA to Protein. In a cell, the DNA code is transcribed to RNA. RNA molecules have different functions. Some protein-coding RNA molecules serve as templates for synthesizing proteins in the process called translation, whereas others serve in regulating the expression of genes from the DNA. Proteins and DNA or RNA need to bind to each other in nearly every important event inside living cells that drives the process of the central dogma. Often, experimental conditions and treatments change how proteins bind to the DNA or RNA. However, with limited known studies, it is not possible to design tools to predict with high confidence how proteins with DNA and RNA binding properties interact with these molecules. If we could predict protein binding to DNA or RNA, we could design better biotechnology-driven applications for improving the expression of genes with desired outcomes, such as those that improve response to stressors, development, and environmental conditions. This collaborative project combines artificial intelligence (AI) tools with biophysical principles to develop software that predicts protein binding to DNA or RNA. This project will use experimental measurements to guide the software development. The developed tools will be available to the public. The project plans to organize educational activities that will introduce AI methods to students at all levels, including K–12, undergraduate, and graduate students. Exposing students to AI methods will prepare them for the future workforce. Protein–nucleic acid interactions underlie essential biological processes, including transcription, translation, DNA repair, and gene regulation. Accurate characterization of these interactions will advance mechanistic understanding of molecular recognition and accelerate the development of diagnostic and therapeutic tools. However, current predictive models usually rely on large training datasets, limiting their accuracy and transferability when applied to new protein-nucleic acid systems. This project will develop a data-efficient, physics-informed AI framework that integrates biomolecular sequence, structural, and dynamic information to predict sequence-specific protein–nucleic-acid binding affinities. The project has three integrated objectives: (1) to develop predictive models for protein–nucleic acid binding under sparse-data conditions; (2) to apply these models to design oligonucleotide aptamers against selected target proteins; (3) to incorporate the validated models into a publicly accessible computational platform for rapid prediction of binding affinities and high-affinity nucleic acid motifs. Model development will be conducted in collaboration with two experimental laboratories with expertise in DNA mismatch recognition and oligonucleotide aptamer selection. Predictions will be evaluated using public databases and new experimental measurements from complementary biochemical and biophysical techniques. This collaborative effort will yield a reliable, generalizable predictive framework that advances our understanding of protein–nucleic acid recognition and enables practical applications in therapeutic oligonucleotide discovery. The project contributes to the national priorities of artificial intelligence and biotechnology. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.