近红外光谱技术在中药质量控制应用中的化学计量学建模：综述和展望

摘要

Near infrared spectroscopy (NIR) is currently the most widely used Process Analytical Technology (PAT) in the pharmaceutical industrial. It's application in the quality control of Chinese herbal medicines has also attracted much attention in recent years. Chinese herbal medicines are often very complex in composition,and the production processes are distinctive from that for pharmaceutical chemicals. As a result,the use of NIR in the Chinese herbal medicine domain faces major challenges, in particular in chemometric model development. This paper provides a review of chemometric methods for NIR model development with the focus on NIR application in on-line quality control in the production of Chinese herbal medicines. The topics include calibration data collection,pre-processing, selection of data for model development and validation,and wave number selection,as well as model building and performance assessment. It is emphasized that construction of models with good performance should be an iterative process integrating spectral pre-processing (e.g. smoothing, derivative method,standardized algorithm,data enhancement algorithm,and principal component analysis),wave number selection (e.g. genetic algorithm,random frog) and linear and nonlinear model construction algorithms (e.g. partial least squares,neural networks and support vector machines). A future platform is envisaged as one that shares data and integrates sophisticated algorithms in the background and presents users with friendly,easy to use,intelligent and semi-automated interfaces. Review and discussions has been made based on practical case studies.%近红外光谱（NIR）是制药工业领域应用最为广泛的过程分析技术（PAT），在中药产品质量的在线实时检测和控制中越来越受到重视。和化学药相比，由于中药组成的复杂性和生产加工过程的特殊性，对利用化学计量学建立NIR预测模型，提出了新的挑战。本文对NIR在中药质量控制应用中的化学计量学建模方法和技术进行了综述并对未来发展做了展望。综述涉及到 NIR 数据的采集、预处理、分组，特征波段自动选取，建模以及模型的验证和评价。讨论了平滑、导数、标准化算法、数据增强算法和主元分析等预处理方法对模型影响。特征波段的选取述及间隔偏最小二乘、遗传算法、无信息变量消除、随机蛙跳法、竞争自适应重加权采样和重要变量投影法等；建模方法论及线性和非线性技术包括主元回归、偏最小二乘回归、人工神经网络和支持向量机回归等。未来的 NIR 建模平台应该是一个在后台集成各种复杂的数学算法和实现数据的无缝共享，面向用户的前台则是友好、简单、智能的半自动界面环境。论述结合具体的实例进行。