摘要
A reliable 3-D structure of Triacylglycerol lipase from Bacillus thermocatenulanatus was constructed by homology modeling. Under molecular dynamics simulation, it was refined and checked. The model was further used as a receptor to search binding sites and carry out flexible docking with a range of substrates, whose enzyme activities were already measured. By inputting a series of docking results, virtual substrates screening models were established and assessed. Monadic nonlinear solution demanded less data but was weak in fitting enzyme activity data with little difference; its mean absolute percentage error （MAPE） of regression was 0.67 and mean square error （MSE） was 1.73 U/mg. Both quadratic stepwise regression and BP neural network were good in regression and prediction; however, more data were required. In the cross-validation of 12 groups, overall MAPE of regression and prediction for the former were 0.18 and 0.49, while the latter was 0.55 and 0.36. MSE values for these two methods were 0.95 and 1.20 U/mg, respectively. Therefore, monadic nonlinear regression model can be used as a preliminary screening one; quadratic stepwise regression and BP neural network approach can be applied to precise screening.
A reliable 3-D structure of Triacylglycerol lipase from Bacillus thermocatenulanatus was constructed by homology modeling. Under molecular dynamics simulation, it was refined and checked. The model was further used as a receptor to search binding sites and carry out flexible docking with a range of substrates, whose enzyme activities were already measured. By inputting a series of docking results, virtual substrates screening models were established and assessed. Monadic nonlinear solution demanded less data but was weak in fitting enzyme activity data with little difference; its mean absolute percentage error （MAPE） of regression was 0.67 and mean square error （MSE） was 1.73 U/mg. Both quadratic stepwise regression and BP neural network were good in regression and prediction; however, more data were required. In the cross-validation of 12 groups, overall MAPE of regression and prediction for the former were 0.18 and 0.49, while the latter was 0.55 and 0.36. MSE values for these two methods were 0.95 and 1.20 U/mg, respectively. Therefore, monadic nonlinear regression model can be used as a preliminary screening one; quadratic stepwise regression and BP neural network approach can be applied to precise screening.
基金
Supported by the Research Found for Doctoral Foundation of Institutions of Higher Education of China (20070385001)
