ИСТИНА

Recently, QSPR modeling of logK for the 1:1 complexation of 537 organic HBAs with 4-fluoro-phenol as the reference HBD was peformed applying new ISIDA fragment descriptors able to describe “local” properties related to selected atoms or functional chemical groups [2]. The logK values for the modeling were measured by Fourier transform IR spectroscopy (FTIR) in CCl 4 at 298 K and experimentally attributed to HBA sites taking part in the complexation. Unlike previous QSPR studies of H-bond complexation, the models based on the FTIR data and new ISIDA descriptors are able to predict the H-bond strength (basicity) of the O, N and S atoms in different hybridization states as HBAs of the polyfunctional molecules. QSPR models were obtained using support vector machine (SVM) and ensemble multiple linear regression (ISIDA MLR) methods. The models were validated with cross-validation procedures and with two external test sets. The root mean squared error of the 5-fold cross validation ISIDA MLR predictions is 0.24 of logK units. For predictions of the HBA strength of organic molecules, the SVM consensus model is freely available on web server: http://infochim.u-strasbg.fr/webserv/VSEngine.html.