Data-Driven Exploration of Silver Nanoplate Formation in Multidimensional Chemical Design Spaces

Artificial Intelligence (AI) driven closed systems, which are usually composed of an AI agent to plan experiments, robots to perform experiments, and a high throughput characterization method to evaluate experiments have recently shown to be successful in optimizing structural properties of colloidal nanoparticles. However, a limitation of these kinds of systems is that they have only been demonstrated in constrained design spaces which is where the targeted structure has a high probability of being formed. In addition, while many samples are being synthesized and characterized, minimal amounts of information on the relationship between experimental design parameters and the structure of the nanoparticles is obtained from the experiment. To solve these problems, we introduce a novel AI driven closed system and test it with a model system of silver nanoparticle synthesis with the objective of synthesizing nanoplates. Our method first searches the design space for silver nanoplates based on UV-Vis spectroscopy curves, which are autonomously classified into “plates” or “not plates” using a distance metric. This information is then used to train a gaussian process classifier which then iteratively suggests experimental parameters for a new batch of samples that are likely to be nanoplates. After the chemical design space is constrained to contain mostly nanoplates, we then use small angle x-ray scattering characterization to obtain size/shape parameters. This information is used to train a gaussian process regressor, from which we can extract design rules such as the effect of the composition of the reagents on the obtained size/shape parameters.