Communications in Information and Systems

Volume 18 (2018)

Number 4

Prediction of molecular energy using deep tensor neural networks

Pages: 229 – 250

DOI: https://dx.doi.org/10.4310/CIS.2018.v18.n4.a2

Authors

Yan Li (State Key Laboratory of Molecular Reaction Dynamics, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian, Liaoning, China)

Guohui Li (State Key Laboratory of Molecular Reaction Dynamics, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian, Liaoning, China)

Hanyi Min (Chinese Academy of Medical Science and Peking Union Medical College Hospital, Ophthalmology, Beijing, China)

Zibing Dong (School of Physics and Electronic Technology, Liaoning Normal University, Dalian, Liaoning, China)

Tian Yuan (School of Physics and Electronic Technology, Liaoning Normal University, Dalian, Liaoning, China)

Xiaoqi Li (School of Physics and Electronic Technology, Liaoning Normal University, Dalian, Liaoning, China)

Peijun Xu (School of Physics and Electronic Technology, Liaoning Normal University, Dalian, Liaoning, China)

Abstract

In this paper, we propose a combined scheme called Quantum Mechanics and Deep Tensor Neural Network (QM-DTNN) to address the challenges related to the prompt and accurate calculation of the physicochemical properties of a protein. In QM-DTNN, a protein is decomposed into individual amino acid units that are treated with molecular caps. The physicochemical properties (molecular energy) of amino acid units are predicted using DTNNs, which are trained by QM (ab Initio) data. The training, validating, and testing data sets are made of conformations drawn through enhanced sampling in specific collective variable spaces. The inputs of DTNNs include pair-wise inter-distance and nuclear charges of an amino acid unit. The outputs of DTNNs, which are the physicochemical properties of amino acid units, are calculated using QM. The three typical amino acid units (i.e., Arginine, Lysine, and Tryptophan) are used to demonstrate the feasibility of QM-DTNN. The prediction results demonstrated good correlations to QM data. The proposed scheme reduces the computational time considerably compared to that of QM calculation with acceptable precision loss.

Full Text (PDF format)

Published 26 November 2018