ИСТИНА

It is widely known that for evaluation of compound's efficacy it is necessary to carry out a comprehensive assessment, taking into consideration different criteria mostly related to potency, pharmacokinetics, and safety. This allows reaching the balance between different properties, which ensures the way for successful drug design. Such approach is an example of multiparameter optimization (MPO). At the same time it is obvious that compound properties, both experimental and predicted, are usually characterized by large errors and it is not always possible to fix this problem. In the present work we investigated to what extent the contribution of calculation errors of single or multiple properties is significant in the course of MPO. We chose the 'indices' as the analyzed quantitative compound parameters which can be calculated based on desirability algorithm using the StarDrop platform. Desirability provides a simple, effective approach to MPO. It describes multiple descriptors of compounds measured on different scales as desirability functions. These are then integrated into a single score. As combinations of individual desirability functions we utilized quantitative estimate of drug-likeness (QED) and two common indices necessary for the development of intravenous or oral drugs for the central nervous system. All data comparison was carried out using several databases: GSK-3 inhibitors, drug-like and random compounds from ZINC, Commercial Compound Collection, and ZINClick databases. In all cases descriptors for desirability functions are based on prediction methods. The investigation revealed that compounds characterized by low drug-likeness, thus lying outside of applicability domains of common models, cannot be evaluated satisfactorily. The requirement for new models for property prediction outside of the “Lipinski rule space” is highlighted by these data. The observed relationships between the descriptors features and the index values are presented.