Predicting DNA Origami Stability in Physiological Media by Machine Learning

DNA origami nanostructures offer substantial potential as programable, biocompatible platforms for drug delivery and diagnostics. However, their structural instability under physiological conditions remains a major barrier to practical applications. Stability assessment of DNA origami nanostructures has traditionally relied on image-based and empirical approaches, which are time-consuming and difficult to generalize across conditions. Here, a proof-of-concept framework coupling dynamic light scattering (DLS) with machine learning (ML) to estimate diffusion coefficient-based stability responses is presented. We use DLS as a screening proxy, supported by gel electrophoresis and atomic force microscopy (AFM) for selected conditions. A dataset of over 1400 measurements across three DNA origami shapes is assembled under physiologically relevant variations in temperature, incubation time, MgCl₂ concentration, pH, and DNase I concentration. The dataset is used to train a consensus ML model built from Gaussian Process Regressor (GPR) and Random Forest (RF) capable of estimating diffusion coefficients for new condition combinations within and near explored ranges. The dataset and ML model are provided as a resource for the community, enabling others to extend and refine stability prediction for diverse nanostructures and conditions. This work establishes a scalable, data-driven framework for guiding the rational design of robust DNA origami nanostructures for biomedical applications. While qualitative and shape-dependent, the framework and dataset provide a scalable basis for community benchmarking and extension.