Nanostructured Au-Based Surface-Enhanced Raman Scattering Substrates and Multivariate Regression for pH Sensing
Nanostructured Au-Based Surface-Enhanced Raman Scattering Substrates and Multivariate Regression for pH Sensing
复制标题
DOI：
10.1021/acsanm.1c00549
复制
发表时间：
2021-06-15
影响因子：
5.9
通讯作者：
Vikesland, Peter J.
中科院分区：
文献类型：
--
作者：
Kang, Seju;Nam, Wonil;Vikesland, Peter J.
一系列介质的兼容性对于表面增强拉曼散射 (SERS) 支持的 pH 检测至关重要。我们报告了使用自上而下的纳米结构金 SERS 基底和多元回归在一系列介质中进行通用 pH 检测。具有垂直堆叠的多个纳米间隙热点的 SERS 基底，用传感分子 4-巯基吡啶 (4-Mpy) 功能化，表现出高空间均匀性。标准比例 pH 检测能够开发基于玻尔兹曼方程的磷酸盐缓冲盐水校准曲线。然而，该校准曲线不能用于预测其他介质（例如碳酸盐缓冲液、苹果汁、牛奶和废水）中的 pH 值。为了解决这些不同介质成分中出现的 SERS 干扰，多元回归已成功应用于所有五种介质的 pH 预测。 4-Mpy SERS 光谱中总共提取了 19 个光谱特征并用于模型开发。在其他多元回归模型中，具有 5/2 Matern 核函数的非参数高斯过程回归模型表现出最高的 pH 预测精度，均方根误差为 0.81。该模型具有通用性，能够确定未用于模型训练的介质内的 pH 值。
Compatibility in a range of media is vitally important for surface-enhanced Raman scattering (SERS)-enabled pH detection. We report universal pH detection in a range of media using top-down nanostructured gold SERS substrates and multivariate regression. SERS substrates with vertically stacked multiple nanogap hotspots functionalized with the sensing molecule 4-mercaptopyridine (4-Mpy) exhibited high spatial uniformity. Standard ratiometric pH detection enabled development of a Boltzmann equation-based calibration curve for phosphate-buffered saline. This calibration curve, however, could not be used to predict pH in other media such as carbonate buffer, apple juice, milk, and wastewater. To address SERS interferences that occur in these different media compositions, multivariate regression was successfully applied to pH prediction for all five media. A total of 19 spectral features in the 4-Mpy SERS spectra was extracted and used for model development. A nonparametric Gaussian process regression model with a 5/2 Matern kernel function exhibited the greatest pH prediction accuracy with a root-mean-square error of 0.81 among other multivariate regression models. This model was generalizable and capable of determining pH within media that had not been used for model training.