ИСТИНА |
Войти в систему Регистрация |
|
ИПМех РАН |
||
The blood-brain barrier permeability of drugs and other physiologically active compounds is among the key pharmacokinetic properties that determine their bioavailability in the central nervous system and critically affect the efficacy, pharmacological profile, mode of use, and safety of drugs. In the design of neuroactive compounds, ensuring the optimal blood-brain barrier permeability is an important objective that needs to be balanced with their activity and other properties. Thus, the efficiency of such optimization can be significantly improved if a predictive QSAR model could not only estimate the permeability value but also identify specific structural features responsible for increasing or decreasing it. Using the fragmental descriptors and deep neural networks, we have built an ensemble model of the relationship between the structure of organic compounds and their ability to penetrate the blood-brain barrier that is superior in the prediction accuracy and the applicability domain to the models previously published in the literature. To build the permeability map for a particular compound, first the partial derivatives of the predicted value with respect to each fragmental descriptor are calculated. These derivatives represent the coefficients of the local quasi-linear model. For each atom in a structure, an effective permeability contribution is determined as a sum of coefficients (divided by the fragment sizes) for all fragments it belongs to. The significance of the resulting contributions is evaluated taking into account the distribution of their values in a representative training set of compounds. The significant contributions are presented graphically as a structural permeability map that can be used to guide the design and optimization of neuroprotective, cognition enhancer, or other neuroactive compounds.