We propose a machine-learning model, based on the random-forest method, to predict CO adsorption in thiolate protected nanoclusters. Two phases of feature selection and training, based initially on the Au 25 nanocluster, are utilized in our model. One advantage to a machinelearning approach is that correlations in defined features disentangle relationships among the various structural parameters. For example, in Au 25 , we find that features based on the distribution of Ag atoms relative to the CO adsorption site are the most important in predicting adsorption energies. Our machine-learning model is easily extended to other Au-based nanoclusters, and we demonstrate predictions about CO adsorption on Ag-alloyed Au 36 and Au 133 nanoclusters.

Doping metal nanoclusters with a second type of metal is a powerful method for tuning the physicochemical properties of nanoclusters at the atomic level and it also provides opportunities for a fundamental understanding of alloying rules as well as new applications. Herein, we have devised a new, one-phase strategy for achieving heavy Ag-doping in Au(SR) nanoclusters. This strategy overcomes the light doping of silver by previous methods. X-ray crystallography together with ESI-MS determined the composition of the product to be [AgAu(SCH)] with x ∼ 21. Cryogenic optical spectroscopy (80-300 K) revealed fine features in optical absorption peaks. Interestingly, the heavy doping of silver does not significantly change the electron-phonon coupling strength and the surface phonon frequency. DFT simulations reproduced the experimentally observed trend of electronic structure evolution with Ag doping. We further investigated the electrocatalytic performance of such heavily Ag-doped nanoclusters for oxygen reduction in alkaline solutions. The mass activity of ligand-off [AgAu(SCH)] nanoclusters (217.4 A g) was determined to be higher than that of ligand-on nanoclusters (29.6 A g) at a potential of -0.3 V (vs. Ag/AgCl). The rotating disk electrode (RDE) studies revealed the tunable kinetic features of the nanoclusters by ligand removal.

Determining the aqueous solubility of molecules is a vital step in many pharmaceutical, environmental, and energy storage applications. Despite efforts made over decades, there are still challenges associated with developing a solubility prediction model with satisfactory accuracy for many of these applications. The goals of this study are to assess current deep learning methods for solubility prediction, develop a general model capable of predicting the solubility of a broad range of organic molecules, and to understand the impact of data properties, molecular representation, and modeling architecture on predictive performance. Using the largest currently available solubility data set, we implement deep learning-based models to predict solubility from the molecular structure and explore several different molecular representations including molecular descriptors, simplified molecular-input line-entry system strings, molecular graphs, and three-dimensional atomic coordinates using four different neural network architectures—fully connected neural networks, recurrent neural networks, graph neural networks (GNNs), and SchNet. We find that models using molecular descriptors achieve the best performance, with GNN models also achieving good performance. We perform extensive error analysis to understand the molecular properties that influence model performance, perform feature analysis to understand which information about the molecular structure is most valuable for prediction, and perform a transfer learning and data size study to understand the impact of data availability on model performance.