This study employed four distinct modeling techniques to analyze the structure-activity relationships (SAR) of 13 highly substituted piperidine derivatives across five concentration levels. Among the linear and nonlinear methods evaluated, the hybrid approach combining a Genetic Algorithm with an Artificial Neural Network (GA-ANN) and other machine learning techniques demonstrated better predictive performance, as measured by the correlation coefficient (R2) and root mean square error (RMSE). The study employed a multifaceted computational approach to investigate radical scavenging activity. The Genetic Algorithm-Artificial Neural Network (GA-ANN) method identified several significant molecular descriptors, including the negative logarithm of the diphenylpicrylhydrazyl (DPPH) concentration (–logDPPH), atomic van der Waals volume-weighted parameters, information content related to distance magnitude, and the mass-weighted distance matrix. Furthermore, machine learning analysis confirmed the importance of specific descriptors, emphasizing the mean information content on the distance degree equality, atomic van der Waals volume, and atomic mass-weighted properties as the most relevant for predicting antioxidant activity. Molecular docking studies provided further insights into binding interactions. Among the piperidine derivatives analyzed, derivative 8 exhibited the lowest binding affinity (highest energy) with the 6lu7A receptor, while derivative 3 formed the highest number of hydrogen bonds with the same target. In conclusion, our results suggest that the integrated application of GA-ANN, machine learning, and molecular docking could be a viable strategy. This combined methodology enhances the understanding of the relationship between physicochemical molecular features and biological activity mechanisms, thereby facilitating the rational design of novel therapeutic agents.
Sayyadikordabadi et al. (Wed,) studied this question.