A “targeted” QSPR for prediction of properties

Abstract

Abstract In order to improve the reliability of the Quantitative Structure-Property Relationships (QSPR) for property prediction, a “targeted” QSPR (TQSPR) method is developed, from a training set, which contains only compounds structurally similar to the target compound. Structural similarity is measured by the partial correlation coefficients between the vectors of the molecular descriptors of the target compound and those of the predictive compounds. The available properties of the compounds in the training set are then used in the usual manner for predicting the properties of the target and the rest of the compounds of unknown properties in the set. Preliminary results show that the targeted QSPR method yields predictions within the experimental error level for compounds well represented in the database and fairly accurate estimates for complex compounds that are sparsely represented. The cut-off value of the partial correlation coefficient provides an indication of the expected prediction error.