Uninformative variable elimination assisted by Gram–Schmidt Orthogonalization/successive projection algorithm for descriptor selection in QSAR

摘要: Abstract Employment of Uninformative Variable Elimination (UVE) as a robust variable selection method is reported in this study. Each regression coefficient represents the contribution corresponding established model, but presence uninformative variables well collinearity reliability coefficient's magnitude suspicious. Successive Projection Algorithm (SPA) and Gram–Schmidt Orthogonalization (GSO) were implemented pre-selection technique for removing redundancy among model. elimination-partial least squares (UVE-PLS) was performed on pre-selected data set C value 's calculated each descriptor. In case UVE assisted by SPA or GSO could be used order to rank according their importance. Leave-many-out cross-validation (LMO-CV) applied ordered descriptors selecting optimal number descriptors. Selwood including 31 molecules 53 descriptors, anti-HIV 107 160 utilized When set, obtained results desired not only prediction ability constructed model also selected informative By applying GSO-UVE-PLS data, an optimized condition, seven out with q 2  = 0.769 R  = 0.915. Also SPA-UVE-PLS nine  = 0.81,  = 0.84 Q F3  = 0.8.