Following the large number of descriptors used in QSAR/QSPR, it has become a bottleneck problem how to choose the descriptor set which can be used to develop a good stable and predictable model. In this work, the partial least squares (PLS) method was used to screen the important descriptors. The 42 molecular descriptors were selected from an original pool of 1664 descriptors of 63 organic compounds. The PLS regression model between 42 descriptors and the logarithm of the permeability coefficients of various organic compounds through low-density polyethylene was developed and validated by the variable selection and modeling based on prediction (VSMP) technique. It has been found that PLS regression model has good quality, r²＝0.9647 and q²＝0.8364 for the training set of 43 samples and ＝0.9306 for the test set of 20 compounds. Using PLS variable selection procedure, it is possible to rapidly and effectively select the important variables closely related with the activity of compounds and construct a model with good stability and predictability.

Key words: variable selection, partial least squares (PLS), variable importance in projection (VIP), QSAR, permeability coefficient