QSAR Study On Angiotensin I-converting Enzyme Inhibitory Peptides

In recent years the incidence of hypertension has been rising, and it has resulted inheavy burden on society and family. Angiotensin I-converting enzyme(ACE) inhibitorypeptide from food protein can inhibit ACE activity in vivo significantly,but the specificmechanism is unclear.Quantitative structure activity relationship(QSAR) research isuseful to establish quantitative model between the molecular structure and activity.QSAR research on ACE inhibitory peptide is helpful for looking into action mechanismof ACE inhibitory peptide,and then guiding the design and development of effectivehypotensor.SVHEHS descriptor was derived from457physicochemical properties indexes of20natural amino acids,and its performance was tested by58angiotensin I-convertingenzyme inhibitory dipeptides and48bitter taste dipeptides.The correlative coefficientR~2,root mean square error RMSE,the cross-validation correlative coefficient Q~2LOOandthe external validation correlation coefficient Q~2extof two models were0.936,0.259,0.854,0.737;0.949,0.136,0.886and0.543,respectively.Compared the above statisticsparameters of models with other models constructed by various descriptors,the fittingability,robustness and forecasting ability of SVHEHS were better.These showed that thedescriptor can be used for QSAR research.SVHEHS descriptor and multiple linear regression algorithm were used to studythe QSAR of angiotensin I-converting enzyme inhibitory dipeptides,tripeptides andtetrapeptides.The parameters R~2, RMSE, Q~2LOO, Q~2extof dipeptides model were0.851,0.327,0.781and0.792,respectively.Hydrophobicity and charge of C-terminal aminoacid,and steric properties of N-terminal amino acid were closely related with its activity.Hydrophobicity of C-terminal amino acid was most positively associated with theactivity,and charge of C-terminal amino acid most negatively with the activity.Theparameters R~2, RMSE, Q~2LOO, Q~2extof tripeptides model were0.805,0.339,0.717and0.817,respectively.Hydrophobic,electronic and steric properties of N-terminal aminoacid,hydrophobic and electronic properties of C-terminal amino acid were closelyrelated with its activity,especially the hydrophobicity of N-terminal amino acid wassignificant factor with positive effect.The parameters R~2, RMSE, Q~2LOO, Q~2extoftetrapeptides model were0.792,0.393,0.553and0.630,respectively.Hydrogen bondscontributions and electrical characteristics of the third amino acid were closely relatedwith its activity,especially hydrogen bonds contributions was significant factor with negative effect.The parameters R~2, RMSE, Q~2LOO, Q~2extof all of the peptides modelwere0.744,0.508,0.532and0.567,respectively.SVHEHS descriptor and partial least square algorithm were used to study theQSAR of angiotensin I-converting enzyme inhibitory dipeptides, tripeptides andtetrapeptides.The parameters R~2, RMSE, Q~2LOO, Q~2extof dipeptides model were0.607,0.587,0.507and0.783,respectively.Hydrophobic,electronic and steric properties ofC-terminal amino acid,steric features of N-terminal amino acid were closely relatedwith its activity.The parameters R~2, RMSE, Q~2LOO, Q~2extof tripeptides model were0.852,0.232,0.813and0.839,respectively.Hydrophobic and steric properties of N-terminalamino acid, electronic and steric properties of C-terminal amino acid were closelyrelated with its activity.The parameters R~2, RMSE, Q~2LOO, Q~2extof tetrapeptides modelwere1,0,1and0.935,respectively.Electronic properties of the third amino acid,stericproperties of the second amino acid,steric properties of the C-terminal aminoacid,hydrogen bonds contributions of N-terminal amino acid were closely related withits activity.The parameters R~2, RMSE, Q~2LOO, Q~2extof all of the peptides model were0.862,0.356,0.829and0.632,respectively.SVHEHS descriptor and artificial neural networks algorithm were used to studythe QSAR of angiotensin I-converting enzyme inhibitory dipeptides,tripeptides andtetrapeptides.The parameters R~2, RMSE, Q~2LOO, Q~2extof dipeptides model were0.946,0.249,0.951and0.852,respectively.Steric features of N-terminal amino acid andC-terminal amino acid were closely related with its activity.The parameters R~2, RMSE,Q~2LOO, Q~2extof tripeptides model were0.973,0.135,0.945and0.813, respectively.Hydrophobic and steric properties of N-terminal amino acid, steric properties ofC-terminal amino acid were closely related with its activity. The parameters R~2, RMSE,Q~2LOO, Q~2extof tetrapeptides model were0.915,0.250,0.879and0.814, respectively.Steric features of C-terminal amino acid,electronic properties of the second amino acid,electronic properties of the third amino acid were closely related with its activity. Theparameters R~2, RMSE, Q~2LOO, Q~2extof all of the peptides model were0.958,0.224,0.948and0.634, respectively.Finally,advantages and disadvantages of the three algorithms were compared.Multiple linear regression can obtained clear function expressions, and be helpful formodel interpretation and analysis. However, its requirements of variable orthogonalitywas strict. Partial least squares had great advantage to small samples with many independent variables and serious related data, but it could not fit all of the databasewell. Artificial neural network can fit nonlinear relation, and approach the best sampledata law quickly, but it had no explicit function equation. Moreover, there wereuncertain rules for the network layers and nodes, learning and memory of the networkwere less stable than the former two algorithms. Although fitting abilities of threealgorithms to peptide library varied from each other, models drew the common structurefeatures which affect the peptides activities. Hydrophobicity and charge of C-terminalamino acid,steric properties of N-terminal amino acid were more associated withdipeptides activity. Hydrophobic and steric properties of N-terminal amino acid wereassociated with tripeptides activity much. Charge of the third amino acid was associatedwith tetrapeptides activity much.Models, which established by peptides with differentlength, can make a prediction for the activities of different peptides in one model. Hence,it can consume less time for modeling. This method can facilitate activities prediction ofdifferent peptides when the impact factors on activities were clear.