Autonomous retrosynthesis of gold nanoparticles via spectral shape matching

Abstract

Synthesizing complex nanostructures and assemblies in experiments involves careful tuning of design factors to obtain a suitable set of reaction conditions. In this paper, we study the application of Bayesian optimization (BO) to achieve autonomous retrosynthesis of a specific nanoparticle or nano-assembly structure, shape, and size starting from a set of reagents selected a priori. We formulate the BO as a shape matching problem given target spectra as a structural proxy with a goal to minimize the shape discrepancy. The proposed framework is grounded in analyzing the spectra as belonging to function spaces and a Riemannian metric defined on them. The metric decomposes spectral similarity into amplitude and phase components. It provides a shape matching distance to optimize as opposed to purely intensity similarity obtained from the commonly used mean squared error (MSE). Applying the framework to experimental and simulated spectra, we demonstrate the advantage of shape matching over MSE and other generic functional distance measures.