Data-Driven Analysis of Experimental Design Spaces for Colloidal Synthesis and Assembly

Abstract

Colloidal nanomaterials offer diverse functionalities driven by their structural properties, requiring precise control over synthesis and assembly. Due to the complexity of the experimental design space, a self-driving lab approach, which combines artificial intelligence (AI) with autonomous experimentation, is a powerful method used to navigate it. In this approach, an AI agent selects and evaluates experiments in a closed loop, progressively improving its understanding of the design space as data is collected. Some ways to improve self-driving labs include improving the distance metric to enhance the system's ability to guide synthesis. In this work, the amplitude-phase distance metric is introduced, which captures shape differences in functional datasets (e.g., UV-Vis Spectroscopy, SAXS). Its performance is compared to the Euclidean distance metric, and key differences are observed in nanoparticle structural differentiation (e.g., between nanospheres and nanorods) and the balance between the exploitation and exploration of the design space. Further advancements in self-driving labs include using multiple characterization methods (UV-Vis, SAXS, TEM), minimizing reliance on literature for design space definition, and employing interpretable AI to extract experimental insights. These improvements are demonstrated with another self-driving lab tested with silver nanoplate synthesis, where the derived design rules align with established knowledge. In addition to inorganic nanoparticle synthesis, this work also explores engineering stimuli responsive protein assemblies, which was done by modifying the RhuA protein with light and chemically responsive molecules. Structural analysis via a Monte Carlo-based SAXS fitting method reveals light-controlled assembly into tubes or sheets, influenced by solution ionic strength. This efficient modeling approach supports future exploration of metastable structures using in situ SAXS combined with AI-guided light sequencing. Finally, we explore DNA-mediated assembly of lipid encapsulated nanoparticles where high-throughput experimentation is used to identify the effect of design variables on the final assembly.