Projection Pursuit Regression Model for Predicting Biodegradability of Substituted Aromatic Hydrocarbon

Weighted holistic invariant molecular(WHIM) descriptors of 42 substituted aromatic hydrocarbons were calculated.The quantitative structure-biodegradability relationship study was performed with biodegradability parameter BOD by multiple stepwise linear regression analysis method and projection pursuit regression technique.The analysis results of stepwise linear regression indicated that the biodegradability of substituted aromatic hydrocarbon was related with molecular shape,atom density distribution and symmetry,and influenced by atomic masses,van der Waals volume and electrotopological indices.The WHIM descriptors can capture the relevant molecular 3D information regarding the molecular size,shape,symmetry,and atom distribution.The projection pursuit regression model for biodegradability of substituted aromatic hydrocarbon was established based on multiple stepwise linear regression results.By using the new idea of "examination of data-simulation-forecast",the model stability and forecast ability were obviously elevated.The traditional correlation coefficient(R2) of projection pursuit regression model was 0.642 and the average percentage error for the testing set was 8.29%.