| Near-infrared spectroscopy(NIRS) is a new developing analysis that combines spectral measurement with chemometrics into a whole. The NIR technology has been extensively applied in agriculture, food, tobacco, environment, biology-medicine analysis. In recent year, NIR technology was used in pharmaceutical research and application, which focus on the study of pharmaceutical raw materials, analysis of pharmaceutical preparations, control of production process of drug, comparison of the predict effects of spectral data through different chemometric methods.This paper main objective is to determine the content of pharmaceutical raw materials and compare different chemometric methods. Firstly, spectral measurement and partial least squares (PLS) chemometric method were used to determine the contents of atomoxetine hydrochloride, fudosteine, dipyridamole and aspirin. The best PLS model of four kinds of components were established in their optimal wavelength range. R, RMSECV, RMSEC and RMSEP were used to evaluate the PLS models of four components. The results of PLS model of atomoxetine hydrochloride were below, R was 0.98468, RMSECV was 0.00290, RMSEC was 0.00117, RMSEP was 0.00476, Factor was 8. The results of PLS model of fudosteine were followed, R was 0.99477, RMSECV was 0.00357, RMSEC was 0.00304, RMSEP was 0.00389, Factor was 3. The results of PLS model of dipyridamole were below, R was 0.99921, RMSECV was 0.00170, RMSEC was 0.00110, RMSEP was 0.00291, Factor was 5. Results of PLS model of aspirin were followed, R was 0.99517, RMSECV was 0.000810, RMSEC was 0.000378, RMSEP was 0.000831, Factor was 8. The optimal wavelength ranges of four components of atomoxetine hydrochloride, fudosteine, dipyridamole and aspirin were as followed,4550cm-1～4950cm-1,4900cm-1～5200cm-1, 4250cm-1～4950cm-1 and 4950cm-1～5250cm-1. The average recovery rates of PLS model of four components were followed,103.9%,99.6%,99.9%and 99.3%. The range of relative error of PLS model of four components ordered by-6.8%～10.2%,-2.0%～1.0%,-2.0%～1.2%,-5.5%-8.3%.Secondly, the optimal wavelength rangs of four components of three drugs would used to combine with BP feed-forward artificial neural network to forecast the contents of four components of three drugs. The predict results were as followed, the predict effects of PLS-ANN model made by raw spectrum were better than those of PLS-ANN model made by first and second derivative spectrum. The R of raw spectrum PLS-ANN model of four components were followed 0.6591,0.8359,0.8646 and 0.4029. RMSECV were followed 0.0318,0.0317,0.0319 and 0.0318. RMSEP were followed 0.0232,0.0194,0.0212 and 0.0277. The average recovery rates were followed 93.9%,101.4%,94.8%and 103.8%. The range of relative error were followed-30.3%～25.5%,-4.8%～8.2%,-9.2%～0.5%,-21.7%～22.1%, respectively.Finally, the R, RMSECV, RMSEP, average recovery rate and average relative error were compared between PLS models and PLS-ANN models of four components of three drugs. Results were followed, both PLS models and PLS-ANN models of four components were able to satisfy the requirements of error range in the actual production. As the ANN method is a relatively new analytical method that development is still not perfect in certain details of the operation. So, the predict effects of PLS method to establish the models of four components is better than those of PLS-ANN models.
|
PDF Full Text Request