The first prediction of protein "optical fingerprint" by artificial intelligence

It is learned from the University of science and technology of China that Professor Jiang Jun of the National Research Center for microscale material science of the University and his collaborators have cooperated to simulate the structure-activity relationship between the structure and properties of protein peptide bonds by using the neural network technology in artificial intelligence machine learning, which greatly reduces the calculation amount and provides an efficient tool for predicting the optical properties of proteins The results were published in the proceedings of the National Academy of Sciences   The spectral response signal of protein, especially the UV spectrum, can be called the "fingerprint" of protein skeleton This "optical fingerprint", through the interpretation of theoretical simulation, can reveal the precise protein structure and provide extremely important information for life science and medical diagnosis   However, the structure of protein is extremely complex and changeable, which requires a lot of high-precision quantum chemical theoretical calculations Even the most powerful supercomputers can't bear it because of the huge amount of computation Therefore, the theoretical interpretation of protein spectrum is a long-term difficulty and challenge, which limits the accurate analysis of spectrum and the discovery of protein structure   At 300K, the researchers first obtained 50000 different peptide bond model molecules by molecular dynamics simulation and quantum chemistry calculation The bond length, bond angle, dihedral angle and charge information are selected as descriptors by machine learning algorithm The structure-activity relationship between the ground state structure of peptide bond and its excited state property is constructed by neural network Based on the trained machine learning model, the ground state dipole moment and excited state properties of peptide bond are predicted Finally, the UV absorption spectrum of peptide bond is predicted In order to verify the robustness of the machine learning model, based on the machine learning model obtained at 300K, researchers predict the UV absorption spectra of peptide bonds at 200K and 400k, and the results are in good agreement with the time density functional theory   This is the first time that artificial intelligence technology has been used to calculate and predict protein spectrum A large number of data are obtained through theoretical calculation After training with artificial intelligence, the feasibility and advantages of machine learning to simulate the UV absorption spectrum of protein peptide bond skeleton are established, and the interpretation of "optical fingerprint" of protein will become more easy and effective.