Model for Parameterization of Interatomic Interaction Potentials by Quantum-Mechanical Descriptors on the Basis of a Graph Neural Network, Модель для параметризації потенціалів міжатомної взаємодії за квантово-механічними дескрипторами на основі графової нейронної мережі

<jats:p>Based on graph neural networks, a machine learning model has been developed to predict the total energy of biomolecules from their structural formulas and quantum-mechanical descriptors by predicting the parameters of the functions that approximate interatomic potentials. The applicability of the created model for predicting the total energy of biomolecules, their interaction energy, and the ordering of conformers by energies has been proven. The physical validity of the obtained parameter values was demonstrated, which opens opportunities for further application of the model in molecular modeling problems.</jats:p>