Comparison of the molecular descriptors efficiency in modeling the structure-activity relationship

Abstract

The purpose of this work is to test the combination of various sets of descriptors in the framework of kNN-QSAR method, including the SiRMS descriptors, which allow interpreting the constructed model. An example of case study and interpretation is shown based on the set, that is consist of 90 nitroaromatic compounds tested for in vivo toxicity. Firstly, was investigated different sets of descriptors generated by the Rcdk package, from which the Simulated Annealing (SA) procedure selects different sets of descriptors, which then utilized to develop regression models. Then the other descriptor generation systems - Dragon 6.0 and SIRMS were used. Important conclusion of this study is that once again confirmed the need to select the appropriate set of descriptors in each particular case, which was many times emphasized by other authors. In addition, this work gives an example of proper interpretation of the constructed models.