Neural Network Modeling of the Multitarget Rage Inhibitory Activity

Abstract—

The models describing the dependence of the level of RAGE-inhibitory activity on the affinity of compounds for target proteins of the RAGE—NF-κB signaling pathway have been constructed by means of the artificial neural networks methodology. For this purpose, a validated database on the structures and activity levels of 183 compounds available in the literature and tested for RAGE inhibitory activity was created. The analysis of AGE—RAGE signaling pathways resulted in detection of 14 key RAGE–NF-κB signal pathway nodes, for which 34 relevant target proteins were identified. A database of 66 valid 3D models of 22 target proteins of the RAGE—NF-κB signal chain was formed. Ensemble molecular docking of 3D models of 183 known RAGE inhibitors into sites of 66 valid 3D models of 22 relevant RAGE target proteins was performed and minimum docking energies for each compound were determined for each target. According to the method of artificial multilayer perceptron neural networks, classification models were constructed to predict the level of RAGE inhibitory activity by the calculated affinity of compounds for significant target proteins of the RAGE–NF-κB signaling chain. The prognostic ability of these models of RAGE-inhibitory activity was evaluated; the maximum accuracy according to ROC analysis was 90% for a high level of activity. The performed analysis of sensitivity of the developed multitarget models revealed most valuable targets of the RAGE–NF-κB signal transduction pathway. It was found that for high level of RAGE inhibitory activity, the most significant biotargets included not only AGE receptors, but eight signaling kinases of the RAGE–NF-κB pathway and the transcription factor NF-κB1. It is suggested that known compounds with high RAGE-inhibitory activity are preferential inhibitors of signal kinases.