Prediction of Photolysis of PCDD/Fs Adsorbed to Spruce [Picea abies (L.) Karst.] Needle Surfaces Under Sunlight Irradiation Based on Projection Pursuit Regression

Quantitative Structure-Property Relationship (QSPR) model for predicting the photolysis half-life (t(1/2)) of PCDD/Fs sorbed to spruce [Picea abies (L.) Karst.] needle surfaces and irradiated by sunlight was firstly developed based on Projection Pursuit Regression (PPR), as a novel machine learning technique, by using the compounds' molecular descriptors calculated from the structure alone. PPR analysis for the PCDDs and PCDFs separately revealed that there is no correlation between them. Three molecular descriptors selected by the Heuristic Method (HM) for the PCDFs were used as inputs to perform Multiple Linear Regression (MLR) and PPR studies. Both linear and nonlinear models gave very satisfactory results: the square of correlation coefficient (R(2)) was 0.828 and 0.893, the Root Mean Square Error (RMSE) was 0.042 and 0.032, respectively, for the whole set. The proposed models can identify and provide some insight into what structural features are related to the log t(1/2) values of compounds and help to improve the understanding for the photolysis mechanism of compounds under sunlight irradiation. Furthermore, this paper provided two new and effective methods for predicting the t(1/2) values of the compounds from their structures and gave some insight into structural features related to the log t(1/2) values of PCDFs.