Machine Learning for the Discovery and Optimization of Organic Dyes

Organic dyes have a variety of applications in modern life. They constitute many of the colors used in textile technology, and have advanced material uses such as dye-synthesized solar cells and organic light emitting diodes (OLEDs). Many of these applications depend on the photonic and electronic structure of the molecule, which can often be understood though DFT and TD-DFT. However the tailored design of dyes for specific applications remains a challenge for scientists and engineers; mainly due to the expense involved in calculating the relevant properties, but also the oftentimes narrow intersection of properties required for tailored applications (a wool dye must me water insoluble, nontoxic, have a net positive charge, be photostable, etc). Here, we present a machine learning based tool for the design and discovery of tailored organic dyes. Machine learning models are used to screen large libraries of molecules and predict their application based on their intersection of properties. A few of the applications included in this tool are textile dye, OLED, food coloring, and photon harvesters. It is our hope that the widespread use of this tool by materials scientists and engineers will allow for an acceleration of dye discovery and a lessening of experimental burden.