Prediction of adverse reactions to antibacterials using MODESLAB methodology

Due to the high interest represented by the study of adverse drug reactions and the importance of anticipating them, the use of molecular modeling methods for this purpose becomes a novel fact. In this work the calculation of the spectral moments of the adjacency matrix between edges of the molecular graph with suppressed hydrogens, weighted on the main diagonal with different parameters that characterize both the bonds and the atoms in the molecules of 63 compounds composed of 63 antibacterial action, using the MODESLAB methodology. 91 descriptors were calculated, which were used in a series of training divided into four groups, according to the most frequent type of adverse reaction. In order to identify the descriptors that best discriminate between the compounds of each group and define the set of functions of these descriptors capable of distinguishing as accurately as possible the members of one or another group, a discriminant analysis was developed using the software Statistical Statistical 8.5. Three functions were generated that constitute linear combinations of 6 molecular descriptors, which encode both steric and electronic information of the molecules of each group. The obtained functions have a very low minimum Wilks Lambda (0.07) and a high canonical correlation (0.82), which demonstrates its discriminant power