QSPR Study Of Gas Chromatography Retention Indices Based On Monte Carlo Model Population Analysis

In the process of gas chromatographic analysis, the qualitativeanalysis of the components is usually achieved by the retentionindices reported in the literature [1]. Hence, the retention indices inthe literature can be used to conduct the qualitative analysis in theabsence of standard samples. Moreover, the retention indices are themacroscopic reflection of microscopic structures of the moleculars.In this dissertation, the main work is to build the QSPR models for thevarieties of organic compounds (naphthenic hydrocarbons,cycloolefins, ethers, amines, ketones, esters, etc.) on two differentstationary phases (squalane and SE-30). Using the softwareADMEWORKS ModelBuilder, the eight descriptors (6th order pathMolecular Connectivity Valence, Electronegativity, FQLogS, Shadowarea2(XZ plane), Shadow area3(YZ plane)，Shadow area1(XY plane)，Mass weighted Width/Thickness and Minimum electron density value)and another seven descriptors (Mass weighted Length, Fractional mass of rotatable atoms, Relative negative charged surface area(based on AM1method), Relative negative charge (based on AM1method), Bond strain energy of molecule, Maximum electron densityvalue and Max.electrophilic superdelocalizability) were served as thestructural parameters of the samples on squalane and SE-30stationary phases, respectively. Then, the correlation analysismodels based on multiple linear regression (MLR), support vectormachine (SVM) and radial basis function (RBF) neural network werebuilt and the gas chromatographic retention indices of theindependent test samples were predicted.In the data analysis, generally a one-time analysis is carried outon one data set, and then a corresponding model is built. Subequently,the model is used in modeling and prediction. However, the model isonly generated from one data set, which results in some limitations tothe model. Considering this problem, the sampling procedure basedon Monte Carlo method was adopted in the dissertation, and then theQSPR models were generated from different data sets. Finally,according to the statistical analysis process, the models based ondifferent modeling tools were evaluated more accurately andunbiasedly.The main work in the dissertation includes:(1)First，the gas chromatographic retention indices of various organic compounds (naphthenic hydrocarbons, cycloolefins, ethers,amines, ketones, esters, etc) on squalane and SE-30stationaryphases were collected and calculated at25℃,30℃,50℃,60℃,64℃,70℃,80℃,100℃,80℃, respectively. Then, the average of theindices was calculated under all the temperatures.(2)This basical theories of the multiple linear regression (MLR),support vector machine (SVM), and radial basis function (RBF) neuralnetwork were introduced briefly.(3)The molecular descriptors of the sample compounds werecalculated using the ADMEWORKS ModelBuilder software.(4)Based on the Monte Carlo Model Population Analysis methods,regression models for the gas chromatographic retention indiceswere built, and the retention indices of the independent test sampleswere predicted by these models.The results of the QSPR models based on Monte Carlo ModelPopulation Analysis were analyzed and compared statistically. Itshows that the SVM models can give more accurate and stable QSPRresults. This indicates that, for dealing with the non-linear data in thisdissertation, support vector machine has its advantages compared with RBF and MLR. Accordingly, SVM can be used as an alternativemodeling tool for QSPR studies.