Research On Enhanced Sampling Of Amino Acids Based On TensorMol

Whether sampling of molecular conformation is sufficient or not in the molecular dynamics simulation directly determines the accuracy of the analysis of the molecular changing process.Different from the traditional sampling methods,scholars have developed enhanced sampling technologies to achieve more adequate sampling of molecular conformation.We can get more states of molecular conformation in a short time by computer simulation.Although there are significant improvements in the efficiency of sampling,the enhanced sampling technology still consumes a lot of computing time and resources.Deep learning is a theoretical framework developed to achieve classification and regression via simulating human thought processes in recent years.Through the training of the samples,the deep learning model can learn the relationship between the different properties of the samples.And this relationship can be reflected through different network structures.Because it implements fast and does not completely rely on the specific physical rules,the deep learning models have been used to solve many problems in the molecular dynamics simulation process.TensorMol is one of the deep learning models that be used to accelerate the speed of MD simulation.In the network structure,TensorMol takes the atomic coordinates and corresponding property parameters in the molecular conformation as the network input.It takes the molecular energy and the corresponding force of the atom as the network output.The network is trained through a large number of conformations.TensorMol applies two subnetworks to achieve the molecular dynamics simulation process based on the consideration of molecular potential energy and long-range interaction between molecules simultaneouly.This paper focuses on the enhanced sampling in the MD process via TensorMol model.First,under different parameters,the properties are compared and analyzed of 10 water molecules.The results shows that the TensorMol deep learning model has good sampling ability.Secondly,based on the traditional AMBER force field and the TensorMol deep learning model,the enhanced sampling process for the two principal dihedral angles is implemented through several amino acid.Through the comparative analysis of the results,the advantages and disadvantages of TensorMol in sampling speed and sampling range are explained.