Transition-Metal-Doped Molybdenum Disulfide Gas Sensors Designed by First Principles and Adsorption Feature-Integrated Machine Learning

Recently, transition-metal-doped molybdenum disulfide (TM-MoS 2 ) has become a frontier in gas sensor research. However, high-quality TM-MoS 2 with both outstanding sensitivity and high selectivity remains scarce, as discovering new TM-MoS 2 is impeded by the insufficiency of robust and rapid predictions of gas adsorption properties of the huge amounts of possible candidates. First-principles calculations provide an effective method to investigate electronic properties and offer a critical reference for experimental preparation, while the high consumption of computational time hinders their broad exploration of all possible material systems. Herein, the present work proposes a novel first-principles machine learning (FPML) approach to directly predict gas sensing trend of TM-MoS 2 and efficiently discover new superior material systems from over 870 datasets. The accuracy and reliability of first-principles calculations for analyzing gas detection performance are demonstrated by experimental results first. By establishing a series of novel feature descriptors, including adsorption energy, band gaps, work function, and charge transfer, requiring low computation cost, the ML model is constructed by high accuracy ( R -square value of 0.95) and high reliability (mean absolute error value of 0.13) for adsorption energy, requiring a reduced amount of training data. We discovered more than 42 promising unreported candidates by the ML model. Some of the systems exhibit higher performance than the reported materials. This work supplies an efficient and effective approach to discover gas detection trends of different two-dimensional metal compounds and other nanomaterials, which can expedite the rational design of the best novel gas sensor for various applications.