QSPR Models on Fragment Descriptors
Vitaly P. Solov’ev, Alexandre A. Varnek · 2017
This chapter illustrates QSPR modeling by the ISIDA_QSPR program, realizing Multiple Linear Regression (MLR) analysis on the base of ISIDA Substructure Molecular Fragment (SMF) descriptors. ISIDA SMF descriptors are counts of the occurrence of subgraphs (fragments) in a molecule, where each descriptor element is associated to one of the detected possible fragments, complying with the user-proposed fragmentation scheme (fragment type, size, etc). The program builds MLR models combining forward and backward stepwise variable selection techniques. The ISIDA_QSPR program is a graphical interface piloting this workflow and supporting graphical analysis of the results linked to the compound structures. It runs under the Windows operating system (ISIDA_QSPR.exe). It is strongly recommended to use of a non-system disk for the ISIDA_QSPR directory. The following exercises are considered in the chapter: individual MLR model, fragment analysis of the individual MLR model, external n-fold cross-validation, consensus modeling, property predictions and virtual screening.