Predicting the Optical Properties of Gold Nanoclusters Using Machine Learning Approach

TL;DR

This paper uses machine learning to predict the optical properties of gold nanoclusters based on synthesis conditions and ligand types.

Contribution

A novel GXBoost-based machine learning model is proposed to predict the emission wavelength of gold nanoclusters with high accuracy.

Findings

The model achieved prediction errors of 1.7%, 1.6%, and 4.9% for different validation sets.

Using thiolated ligands different from GSH resulted in training and test errors of 0.01% and 3%, respectively.

Key variables influencing optical properties were identified using data preparation techniques like One-Hot Encoding.

Abstract

The synthesis of gold nanoclusters (AuNC) is strongly
influenced
by various reaction conditions, and their optical properties are determined
by factors such as the nature of the ligand and the measuring solvent,
among others. To improve the efficiency of the synthesis of metallic
gold nanoclusters with the desired functionality, the application
of machine learning techniques is a smart choice. In this study, a
model based on the GXBoost algorithm is proposed to predict the maximum
emission wavelength of the AuNC emission from a database that includes
more than 200 scientific articles. The validation of the model was
carried out through the comparison of prediction versus experimental
data (not included in the model) and the training and validation data.
The model showed a percentage error of 1.7, 1.6, and 4.9%, respectively,
indicating a reasonable return. Additionally, an independent regression
was performed when the ligand was a thiolated compound different from
GSH, obtaining a training and test error percentage of 0.01 and 3%,
respectively. In addition, critical variables affecting the optical
properties of nanoclusters were explored, and techniques such as One-Hot
Encoding were used to prepare the data. Finally, this work not only
underscores the relevance of AuNCs in modern science highlighted but
also demonstrates the potential of machine learning as a predicting
tool and design of materials in nanoscience, contributing to the optimization
of their properties for future applications.

Figures