Finding environmental-friendly chemical synthesis with AI and high-throughput robotics

Abstract

Recent environmental challenges have resulted in tremendous interest in Green Chemistry, which includes designing chemical products and processes that reduce the use of environmentally harmful substances. Until now, finding new environmental chemical synthesis has mainly been a trial-and-error process, requiring trained expertise and a lot of work. Here, we report a high-throughput process, combining AI techniques and robotic synthesis, allowing us to find a more environmentally friendly way to synthesize an existing material. The model materials in this study are to replace nitrate salts (NO₃⁻), which might be responsible for algae bloom if leaked into open water, with a chloride salt (Cl⁻), a naturally abundant ion, in the synthesis of a metal-organic framework (MOF), Zn-HKUST-1. Our high-throughput process starts with using large language models (LLM)-based literature summary to create a database on the synthesis of Zn-HKUST-1 with NO₃⁻, so that optimized concentrations of Cl⁻ can be suggested. Subsequently, these suggestions are tested with automatic robotic processes, increasing the speed and precision of the experiments, and finding the optimal synthesis condition. Then, by using human verification as a foundation, we developed an AI-based automated classification algorithm to automatically sort the acquired images into crystals and non-crystals, focusing on low-resource settings. We successfully obtained MOF crystals from ZnCl₂ precursors with this process, which proves that our process holds the promise to accelerate the discovery of new Green Chemistry processes.