Machine Learning-Based\nInterpretation of Optical Properties\nof Colloidal Gold with Convolutional Neural Networks

Gold nanoparticles\nare used in a range of applications,\nbut their\nproperties depend on their shape, size, and polydispersity. A quick,\neasy, and accurate characterization of the particles is therefore\nof high importance, especially in flow synthesis settings where continuous\nmonitoring of the characteristics is desired. Our hypothesis was that\nconvolutional neural networks can be used to extract detailed information\nabout structural parameters of gold nanoparticles from their UV–vis\nspectra, and we have shown that this is possible by predicting size\ndistributions from <i>in silico</i> UV–vis spectra\nfor colloidal gold with high accuracy. Here this was done for both\nspherical and rod-shaped gold nanoparticles. We also show that the\naddition of noise makes the prediction of diameter polydispersity\nmore challenging, but the average diameter, and for rods also aspect\nratio distribution, can be accurately predicted even with the highest\nevaluated level of noise. The model structure is promising and worthy\nof implementation to enable predictions beyond <i>in silico</i> generated spectra. The model, for instance, can find application\nin flow synthesis settings to create a machine learning-driven feedback\nloop for automated synthesis.