The Application Research On Homology Model Concept

NIR analytical technique is the fastest growing since the 1990s, which is the most striking spectrum analysis technology. With the rapid development of computer technology, drive the digital analysis instrument and the development of chemical metrology. The NIR method is mainly by building a series of NIR model, realize the rapid screening of the quality of product. Therefore, constructs the prediction ability wide applicable scope of NIR model is crucial.The pros and cons of the near infrared method and the model adopted by the close relationship. A good model not only with the model choice of near infrared spectrum pretreatment method algorithm, more closely associated with modeling training set sample representativeness. Ideal training set of sample should be able to cover forecast samples of the full range of variation, but the sample is not the more the better, too much sample is introduced into the more the chance of error, the less sample and cannot include all components and the characteristics of the background information, therefore must be selected from a large number of samples is suitable for modeling samples. When encountered in the application of near infrared model training set of sample not covered new samples, model predicted results may appear large deviation. At this point, you can through the model updating method to extend the scope of the original general model.To discuss the Homology sample concept in the judgment of the validation and application of NIR model update, in this study, with Cefuroxime Sodium for Injection, Ceftriaxone Sodium for Injection, Cefazolin Sodium for Injection and Cefoperazone Sodium and SulbactamSodium for Injection as an example. First,the analysis of the model of injection, and the original classification, Homology sample judgment when the model meets the modeling training set does not include the new homology sample, the original model that cannot meet the needs of the forecast, need to update the model; Secondly, the modeling sample all band spectrum and training set of sample average spectral similarity coefficient (r1), as the objective index of model needs to be updated, it is generallybelieved as r1<99.0%, the model needs to be updated;Through this study foundthat when the original model between different homology sample, number of spectral distribution, cause the model to certain spectrum fewer homology samples in the prediction, prediction deviation is bigger, this kind of situation is considered a sample distribution is balanced, at this point,98.0%<r1<99.0%, the overall level of prediction model of the sample can be accepted, but the number by supplementing the homology sample method, realize the optimization of the model, to improve the prediction ability; Combined with the actual situation of different samples at the same time, the model update index to validate and refine.When updating the model, correlation coefficient method and the clustering analysis method were used respectively to screening of new samples; Law of correlation coefficient is a new spectrum average spectra of the training set and the original model the r1 values as a starting point, to 0.02% as the sampling interval, up or down to select new spectrum; According to the classification of homology sample at the same time, combined with PCA map, added to the original model of new homology sample, the class or in the homology sample a small number of samples, complete model updating; The method to select sample is convenient, fast, to avoid the tedious sample selection process, improve the efficiency of model updating. Clustering analysis method is the largest distance in clustering analysis diagram about 1/50 new spectrum can be divided into several classes, and then to the original model of random add a certain amount of all kinds of new spectrum, complete model updating; By comparing the two methods of updating the average prediction deviation, found that two methods of updating the predicted results are basically identical, the preliminary judgment method of correlation coefficient can be as a substitute for clustering analysis method, is applied to update sample selection model.In addition, to discuss in homology sample build a quantitative model, the application of cefixime generality, in this study, with dry cefixime suspension agent, cefixime granules, cefixime capsules, cefixime tablets as the sample. Respectively using the correlation coefficient method (model 12) and cluster analysis (model 13, model 14) choice modeling samples; According to the distribution of homology samples in various dosage forms, reasonable choose the number of samples of each kind of homology, combined with the new spectral position in PCA map at the same time, selecting the sample, cefixime general quantitative model is set up; The average prediction deviation are 3.03%,3.04% and 3.03% respectively; When the model was optimized, added to the model prediction deviation is more than 5% of new spectrum, both increase the relatively small number of spectra, homology samples can effectively improve the prediction ability of the model, the updated model to predict the average deviation of 2.63%,2.65% and 2.90% respectively. Through research, modeling for rational selection of samples and model update samples to provide the reference.