近红外光谱的头孢类药品成分分析与模型传递方法

Study on Cefradine Granules Component Analysis and Calibration Transfer Method Based on Near-Infrared Spectroscopy

药品安全与质量监管迫切需要在线、快速、低成本的成分检测技术。近红外光谱技术在检测成本及速度方面具有显著优势,基于近红外光谱的药品成分检测方法,对于提高药品质量监管水平有着十分重要的研究意义和应用价值。在实际应用中,不同光谱仪器由于性能参数不同,测量光谱存在一定差异,很难实现定量校正模型共享。因此,研究不同光谱仪器之间模型传递对于提高分析效率十分重要。针对头孢类药品成分检测的需要,研究了头孢类药品中三种组分定量校正模型,提出了一种基于马尔可夫链(MC)的转换集选择的不同仪器间定量校正模型传递方法。采用两台不同厂家光谱仪器分别测量56份不同批次的头孢拉定颗粒样品,针对样品的三种组分:头孢拉定、头孢氨苄和水分,使用偏最小二乘法(PLS)建立定量校正模型。通过构建概率矩阵,选择合适的转换集,提高模型转换效率及不同仪器得到光谱数据的建模预测精度。实验结果表明,利用该模型转移算法,可利用少量转换集样本实现不同光谱仪器间定量校正模型转移,模型转移前后,定量校正模型对于三种主成分预测相对误差从9.67%, 52.14%和19.25%,分别下降到到4.37%, 31.12%和11.67%。利用该模型传递方法可以有效修正主从仪器光谱差异,实现了不同仪器测量光谱及定量分析模型传递共享。该研究的建模分析与模型传递方法也为药品成分与质量检测提供了技术支撑。

Near-infrared spectroscopy(NIRS)technology has distinct advantages in component detection for its characteristics of high-speed and low-cost,which is essential for the supervision of drug quality and safety.Studying the method of drug component detection based on NIRS technology is significant for improving the level of drug quality supervision.In fact,owing to differences in performance parameters of different spectroscopic instruments,spectra measured are discrepancy,which brings hardship to the realization for quantitative correction models sharing.Therefore,in order to improve analysis efficiency,the calibration transfer method is discussed.In this paper,the establishment of cephalosporins component correction model and calibration transfer method are studied,and a transformation set selection method based on Markov chain(MC)is proposed.Fifty-six samples of cefradine granules in different batches were used.Spectral data were measured by two Fourier spectrometers.For three components of the sample:cefradine,Cefalexin and water,partial least squares(PLS)method was used to establish a quantitative correction model.MC algorithm is used to construct the probability matrix and select the conversion set,which improves the efficiency of model transformation and the prediction accuracy of spectral data.The experimental results show that the quantitative calibration model transfer between different spectroscopic instruments can be realized by using a small number of sample sets.After the model transfer,the relative error of the quantitative calibration model for the three principal components prediction decreases from 9.67%,52.14%,19.25%to 4.37%,31.12%,11.67%,respectively.The spectral differences between master and slave instruments can be corrected effectively,and the transfer and sharing of measurement spectra and quantitative analysis models of different instruments can be realized.The modeling analysis and model transfer methods studied in this paper also provide technical support for drug composition and quality detection.