Abstract
<jats:p>Gold nanoparticles are of broad interest for biomedical applications because their surface properties can be tuned through functionalization by organic ligands. In biological environments, proteins can non-specifically adsorb on nanoparticle surfaces to form a corona that redefines nanoparticle surface properties and governs biological fate. Understanding and controlling the nanoparticle-protein interactions that shape the protein corona is therefore essential, but such interactions are difficult to anticipate. To address this gap, we develop a computational framework that integrates atomistic molecular dynamics simulations and machine learning methods to predict nanoparticle-protein interactions. Inspired by methods that learn protein–protein interaction patterns from molecular surfaces, we develop a unified surface representation for both gold nanoparticles and proteins. We extract nanoparticle and protein surface features and pairwise complementarity features, with ligand-shell dynamics incorporated into the nanoparticle features through molecular dynamics simulations. We then train a classification model to distinguish unbound, moderate, and strong binding behaviors across 264 distinct nanoparticle-protein pairs, allowing the features that govern binding to be learned. The classification model achieves a one-vs-rest receiver-operator-characteristic area under the curve of 0.739 on held-out proteins, demonstrating predictive transfer across diverse systems. We validate the framework by screening candidate nanoparticle binders for a target protein and by systematically evaluating binding for a nanoparticle library against a panel of serum proteins. Candidates predicted to bind strongly form stable interactions in molecular dynamics simulations, whereas predicted non-binders show weak or negligible association. This surface-centric approach captures the physical and chemical basis of molecular recognition, providing a scalable route toward designing gold nanoparticles with tailored protein adsorption profiles and biological properties.</jats:p>