Tunable and scalable production of nanostar particle platforms for diverse applications using an AI-integrated automated synthesis system

Abstract

The tunable optical properties and exceptional electromagnetic field enhancement of nanostar-based plasmonic nanoparticles make them highly promising for a wide array of biomedical applications. However, a great challenge for their widespread use is the time-sensitive nature of the various processes in the nanostar synthesis workflow, which could lead to imprecise control of their homogeneity and high batch-to-batch variability. To address these challenges, we have developed an automated synthesis system with AI capability to reproducibly synthesize large quantities of nanostar particles. This platform uses key synthesis parameters such as reagent volume and reagent addition timing to systematically evaluate how these factors determine the optical properties and SERS enhancement of gold nanostars and bimetallic nanostars. We developed and trained different machine learning (ML) models using nanoparticle characterization data to predict absorbance features and SERS enhancement from synthesis parameters. We compared the performance of five different machine learning models, including artificial neural networks, support vector regression, and several tree-based models, including random forest, extreme gradient boost, and categorical boost. A grid matrix was fed into the final trained models to create a look-up table to synthesize gold nanostars with an absorbance maximum at specific wavelengths, culminating in the reproducible synthesis of desired nanostar platforms with a peak absorbance wavelength of less than 1.2% difference compared to the target peak absorbance. This machine learning-integrated automated nanostar synthesis platform paves the way for more consistent and scalable production to enable the next phase of investigation for nanostar-based technologies and expand the scope of their current biomedical applications.