حسين شيراني

The potential energy surface is considered one of the most fundamental concepts in understan‎ding the chemical an‎d physical behaviors of molecular systems. In drug design, detailed examination of these surfaces plays a foundational role in optimizing drug–protein interactions, enhancing drug selec‎tivity an‎d reducing side effects. Common quantum mechanical computational methods such as Density Functional Theory (DFT), although accurate, are costly an‎d time-consuming for large molecular systems. In this study, a deep neural network potential known as ANI-1x was employed as a machine learning tool to estimate the potential energy surfaces of the molecular structures of two drugs—Levodopa an‎d Resveratrol—used in the treatment of Parkinson’s disease. We demonstrated that the ANI-1x deep neural network model, trained on quantum data, is capable of predicting the potential energy surfaces of the molecular structures of the two studied drugs with high accuracy an‎d at speeds thousan‎ds of times faster than conventional DFT methods. The electronic energies an‎d structural parameters calculated using DFT with the wB97X functional an‎d all Pople basis sets showed that the 6-31G(d) basis set, when used with the wB97X hybrid functional, exhibits behavior similar to ANI-1x. Furthermore, analysis of the vibrational frequencies of the molecule revealed a linear correlation between quantum data an‎d machine learning predictions. The computations performed for molecular structure optimization an‎d extraction of molecular potential energy surfaces using the ANI-1x method required minimal computational resources. Therefore, the approach of using the ANI-1x computational tool in structure-based drug discovery an‎d design studies can serve as a strategic method.

سطح انرژي پتانسيل , يادگيري ماشين , پتانسيل شبكه عصبي عميق , محاسبات كوانتومي , پاركينسون , لوودوپا , رسوراترول

Potential energy surface , machine learning , deep neural network potential , quantum computations , Parkinson’s disease , Levodopa , Resveratrol

Hossein Shirani

Seyed Majid Hashemianzadeh