Method for screening and terminating structure descriptor of activity related model of pollutant quantitative structure

Abstract

The invention discloses a method for screening and terminating a structure descriptor of an activity related model of a pollutant quantitative structure. The method provided by the invention comprises the following steps of: integrating a cross validation correlation coefficient q2 and a model modification correlation coefficient R2adj, establishing a statistical model of a variable subset to obtain a correlation coefficient r2 between an observed value and a model estimation value and obtain a modification correlation coefficient R2adj; subjecting the variable subset of the process above to cross validation to obtain a cross validation correlation coefficient q2 of the model, wherein the cross validation is carried out by means of two methods, i.e. a leave-one-out cross validation and a leave-many-out cross validation; constructing a new parameter QRadj according to a statistical parameter obtained in the process above, wherein the numerical value of the new parameter QRadj of the same system is proportional to the stability of the model and is proportional to the predictive ability. The method for screening and terminating a structure descriptor of an activity related model of a pollutant quantitative structure provided by the invention has the advantages that the relatively high cross validation correlation coefficient q2 of the model can be ensured while avoiding the presence of over-fitting phenomenon through the new standard QRadj, the QSAR (Quantitative Structure Activity Relationship) model variable combination with low r2 value and high q2 value can be prevented from screening, and the stability and the predictive ability of the model are scientifically described.

Description

technical field [0001] The invention relates to a method for screening and terminating structure descriptors of quantitative structure-activity correlation models of pollutants, that is, when establishing a quantitative structure-activity correlation model, a cross-validation method is used to verify internal samples of the model, and a cross-validation correlation coefficient is constructed q 2 Correlation coefficient with model correction R 2 adj the product of QR adj As the termination characterization of model structure descriptor screening, it is used to describe the stability and predictive ability of the model, and judge the pros and cons of the predictive performance of the model. Background technique [0002] The Quantitative Structure and Activity Relationship (QSAR) model of pollutants has been widely used in environmental ecological risk assessment and human health risk assessment of pollutants (Wang Liansheng, Han Shuokui. Quantitative Structure-Activity of...

AI Technical Summary

Problems solved by technology

But it has been found q 2 and RMSEV There are many problems, as pointed out in Golbraikh et al. q 2 It is only a necessary condition for whether the model has the ability to predict, but not a sufficient condition (Golbraikh A., Tropsha A. Beware of q 2 ! J. Mol. Graph. Mod. 2002, 20 (4), 269-276.); Hawkins clearly pointed out that the q 2 Improper use of can lead to overfitting phenomenon; in fact, many subsets of variables have high q 2 value, but the correlation coefficient of the model itself r 2 The value is very low, even close to 0, that is, used alone q 2 or RMSEV It does not explain the stability and predictive ability of the model (Hawkins D. The problem of overfitting. J. Chem. Inf. Comput. Sci. 2004, 44 (1), 1-12.)

Images