Prediction Of Amino Acid Energy Based On Deep Tensor Neural Network Theory

The main theoretical method for studying molecules is simulating molecular changes and physicochemical properties by computer.This mainly includes molecular dynamics(MD)simulation and Quantum Mechanics(QM).MD simulation is based on the force field.It describes the changing nature of atoms during molecular motion.With the development of the computer hardware,especially GPU,the bio-system with millions of atoms can be simulated via molecular dynamics simulation.Limited by the force field,MD simulation can not describe the chemical reaction very well.QM simulates the motion and interaction of atoms by describing the electron clouds around the atoms.This more accurate calculation can be used to study chemical reactions.But it can only handle the system with a small number of atoms.Deep learning is one of the important parts of artificial intelligence.Its main research direction is to improve the performance of specific algorithms through empirical learning based on massive data.Deep learning has made a breakthrough of both the development of the framework and theories.These fields are not well handled by classical methods.These fields include networks,text and image searching,speech recognition,and bioinformatics.At the same time,deep learning has made breakthroughs in understanding quantum systems.This article provides a method to predict the physicochemical properties of amino acids in proteins.This method is based on the quantitative calculation and deep learning theory(Quantum Mechanics-Deep Tensor Neutral Network,QM-DTNN).The proteins are split into individual amino acid units via MFCC method.The process of decomposing amino acids unit is implemented by adding molecular caps.We ensure the completeness of the data set through the enhanced sampling method of Metadynamics.Then we build training,validation,and testing data sets.The input to the DTNN includes the pairwise distance of atoms in the amino acid and the nuclear charge.The output of the DTNN is the physicochemical properties of the amino acid units via QM.In the deep tensor neural network,the input is converted into a set of tensors by the basis function.The output of network is to extract valid information by convolution.In this paper,a variety of amino acids were selected and the energy was predicted by QM-DTNN.The results were well correlated with QM data.The average value of the absolute prediction error and the standard deviation of the absolute prediction error are small.And the overall accuracy is determined to be high.Furthermore,in the chemical environment,the QM-DTNN method can infer the regularity problem of predicting potential atomic energy.Compared to QM calculations within an acceptable errors,our proposed method significantly reduces the computation time.The main purpose of this work is to quickly and accurately calculate the physical and chemical properties of small molecules.Our approach has been proved to be useful in different application areas.As an example of chemical correlation,the amino acids in proteins are selected for research.Our model is used to predict atomic energy in molecules,demonstrating the potential of machine learning and revealing insights into complex quantum chemical systems.