Research On Improved Calculation And Prediction Of Electrostatic Energy Term In DNA Force Field

DNA is an essential biological macromolecule in organisms,which carries the necessary genetic information to synthesize RNA and proteins.Accurate simulation of the structure of DNA molecule helps researchers to study the function of DNA in depth,which has important guidance for drug development and disease treatment.Molecular force field simulation is a commonly used method to explore the molecular structure of DNA.However,most of the popular force fields used for molecular simulation are not accurate enough for the simulation results of interactions between DNA and molecules such as proteins.One of the main reasons is that the calculation of electrostatic energy in DNA molecules is inaccurate,and electrostatic energy plays an important role in the stable structure of DNA molecules.Therefore,improving the calculation accuracy of electrostatic interaction energy can effectively improve the accuracy of DNA molecular simulation.First,in this paper,a single-stranded DNA molecule and a double-stranded DNA molecule are selected as the main research objects.Different molecular cleavage and saturation methods are utilized,respectively.Atomic multipole moments in DNA molecules is calculated based on quantum chemical topology(QCT)theory,and then it is used to calculate electrostatic interaction energy instead of fixed-point charges.It is proved that the method of atomic multipole moments can accurately calculate the atom-atom electrostatic energy.Then,the minimum convergence internuclear distances of 15 interaction types in single-stranded DNA molecules and double-stranded DNA molecules and the relationship between rank of atomic multipole moments and internuclear distance in the convergent region are studied.In this article,the calculation results of the research objects and other different DNA molecules are compared.It is found that they have the same convergence behavior,proving that the method of atomic multipole moments has good transferability,which is of great significance for the next step of constructing the DNA molecular force fields.Although the calculation of electrostatic energy based on atomic multipole moments is more accurate than fixed-point charge,the calculation process is time-consuming.In this paper,for the first time,models for predicting the atom-atom electrostatic interaction energy in DNA molecules are established based on BP neural network and support vector regression in machine learning algorithms,combined with the new method for classifying atom-atom interaction types according to different atomic environments proposed in this article and data sets constructed in the combination with the atomic environments and calculated atom-atom electrostatic energy that varies with internuclear distances.After comparison,the prediction results have high accuracy,which proves the rationality of the classification and prediction methods.At the same time,it is also possible to predict the electrostatic interaction energy that cannot be calculated due to the complex atomic environments.This method ensures the prediction accuracy of electrostatic interaction energy and improves the calculation efficiency,while can be applied to the research of interactions between DNA and molecules such as proteins/metal complexes to achieve the purpose of guiding drug development and disease treatment.