Study On Quantitative Structure-activity Relationship Of Marine Antitumor Active Substances

Tumor is the result of a long period of interaction between genetic susceptibility factors and internal and external environmental carcinogenic factors.Malignant tumors are one of the most serious diseases that endanger human health.The prevention and control of cancer is a serious public health problem facing many countries.For a long time,human beings have devoted themselves to the research and development of cancer drugs,and the development of marine anti-tumor drugs for decades has fully demonstrated its great potential.The Quantitative Structure-Activity relationship（QSAR）looks for a quantitative relationship between the structure and the activity of a series of compounds with known biological activities,and then predicts the activity of new related compounds.It is of guiding significance to assist in clarifying the mechanism of drug action.Objective1.The molecular descriptors of anti-tumor activity are screened out by QSAR,and the theoretical significance is analyzed,which provided the theoretical basis for the development of new drugs and the guidance of the synthesis of new substances.2.The QSAR model is constructed with different methods,and the optimal model is obtained by analyzing and comparing,in order to predict the unknown activity value of the new compound in the range of application of the model.MethodsIn this paper,42 marine Fascaplysin indole bases and 23 marine Meridine analogues are used as the research objects.The molecular structures of the two marine substances are represented by the indicator index and the molecular electric distance vector index,respectively.Multiple linear regression method（MLR）,stepwise regression method,regression tree and support vector machine algorithm（SVM）are used to screen and model the variables.The QSAR model between IC50 and molecular structure of two marine compounds inhibiting tumor cell proliferation activity is constructed.Using the fitting correlation coefficient R² between the experimental and predicted values of the training data set,the Leave-One-Out（LOO）coefficient Q²_LOO,the fitting coefficient between the experimental values of the test set and the predicted values,and the standard deviation of the external prediction(SDEP _ext)and the root-mean-square error prediction（RMSEP）are used to evaluate the reliability,robustness,stability and internal and external predictive ability of the model.By comparing the evaluation parameters of each model,the optimal model is selected,and the unknown activity value of the compound is predicted by the optimal model in the range of application of the model.ResultsEight molecular descriptors,X₃,X₄,X₈,X₉,X₁₀,X₁₃,X₁₆ and X₁₉,are obtained by screening 42 marine fascaplysin indole base substance.The linear model statistics F=3.914,corresponding P=0.004552,indicates that the model has statistical significance.The determination coefficient R²=0.7632 indicates that the model fits well.A nonlinear model based on regression tree is established.The six variables X₃,X₈,X₉,X₁₀,X₁₆ and X₁₉are included in the model.Their contribution to the results of this decision is:X₃ is 4 times,X₈ and X₁₀ are both 2 times,X₉,X₁₆ and X₁₉ are both once.The internal test coefficients R² and Q²_LOO of MLR model are 0.7632 and0.7501,respectively,and the regression tree model R² and Q²_LOO are 0.8078and 0.7985,respectively.The R² of the two models is larger than Q² but not more than 25%.The model external test results show that the MLR model deviates seriously from the experimental value except for the prediction of compound 28.The residuals of the other 9 compounds are（-0.6800¹.0131）and those of the 10 compounds in the regression tree model are（-0.2864⁰.6572）and there are no serious deviations from the experimental data.The SDEP_ext of the two modeling methods are 13.2913and 0.2983 respectively.Stepwise regression method is used to screen out the key molecular descriptors that mainly affected the inhibition of IC50 value of A549 cell line by marine meridine analogues:AATSC5p,GATS3p,BCUTc-1l,SPC-6,minHBa,MLFER_A and MATS1i.Establishing and testing the models,the parameter results show that the R² and Q²_LOO of the six models established by the SVM algorithm are all greater than 0.6,but R² is larger than Q²_LOOOO and more than 25%,which shows obvious overfitting phenomenon.The results show that the parameters R²,Q²_LOO,R²_ext and RMSEP of the linear model established by MLR method are respectively 0.9985,0.8884,0.8792and 0.1243.R²-Q²/Q² is 12.39%,that is,R² is larger than Q²_LOO but not more than 25%,the model has not been fitted.The MLR model was used to predict the pIC50 values of 11 meridine analogues with unknown activity.The distribution of the predicted values was relatively uniform except for compound 22.Conclusion1.The molecular descriptors that mainly affect the activity of marine fascaplysin compounds in inhibiting CDK4 are X₃,X₈,X₉,X₁₀,X₁₆ and X₁₉,in which the coefficients of variables X₃ and X₁₆ are positive,indicating that the descriptors are positively correlated with bioactivity pIC50,and the coefficients of variables X₈,X₉,X₁₀ and X₁₉ are negative,that is to say,the molecular descriptor is negatively correlated with bioactivity pIC50,and the indicator descriptor X₁₉ indicates the position of benzene ring in biphenyl,which is beneficial to increase the activity of the compound when the p-position is connected to the benzene ring.2.The key molecular descriptors that mainly affect the inhibitory effect of marine meridine analogues on the IC50 value of A549 cell line in vitro are AATSC5p、GATS3p、BCUTc-1l、SPC-6、minHBa、MLFER_A and MATS1i.The coefficients of MATS1i descriptors are positive,that is,the presence of descriptor MATS1i can increase the value of anti-tumor activity of meridine analogues.The introduction of MATS1i descriptors for further drug development and synthesis can effectively improve the antitumor properties of the compounds.3.MLR,regression tree and SVM algorithm can be used to build QSAR model.Data mining algorithm,regression tree and SVM are good methods to solve the complex nonlinear relationship between molecular descriptors in QSAR modeling.However,MLR is still the first choice for solving such problems if there is only a simple linear relationship between variables.