Prediction Of Interaction Energy Between Small Organic Molecules Based On Deep Tensor Neural Network Theory

In recent years,the research on organic molecular system,biological macromolecular system and medicine design has entered the era of rapid development.And the study of these systems is inseparable from the analysis of molecular conformation interactions.Although it's effective to use the calculation method based on the quantum chemical theory and the method of molecular dynamics simulation for this study,the former spends too much time and resource and the latter can't describe the molecule changing process in the chemical reaction.Therefore,it has great theoretical and practical significance to predict the physical and chemical properties between molecules quickly and accurately.Combining with the current popular artificial intelligence method,this paper proposes a kind of Deep Tensor Neural Network combined with the calculation method of Quantum chemistry(Quantum Mechanics-Deep Tensor Neutral Network,QM-DTNN).As neural network input,it is composed of two organic small molecular conformation of the nuclear charge vector and interatomic distance matrix,while,the interaction calculated by quantum chemistry can be used as the output of the neural network then training,verify and predict the network.When constructing the data set,a hierarchical generation of conformations is used to fill the entire available space.We apply the method of cross validation to prevent overfitting,when used the deep neural network.In this paper,we trained the molecular interaction energy by the neural network of C4N2H4 and C3NH5,C6H13NO4 and C6H14O2,which in two different conformations.There was a high consistency between the predicted value and the true value calculated by QM.When analyzing the electrostatic interactions and charge transfer of different energy conformations,we found that their conformation interactions mainly occur in hydrogen bonds and dihydrogen bonds which is consistent with the theoretical research,that is,the weak interactions between small organic molecules composed of C,H,O and N are mainly reflected in hydrogen bonds and dihydrogen bonds.For C4N2H4 and C3NH5,the mean value of the absolute prediction error,the standard deviation value of the absolute prediction error,and the linear fitting parameters a and b are close to the standard value.For C6H13NO4 and C6H14O2,the above parameters are consistent with the experimental error.By the QM-DTNN to predict the interaction energy of the two different conformations,and compared with QM it proves that DTNN is feasible in the calculation of molecular interaction energy,and the calculation time is greatly shortened,which provides a feasible method for the study of biomolecular system and medicine design.