定量预测α-淀粉酶及其突变体的酶反应米氏常数(英文)

Quantitative Prediction of Michaelis-Menten Constant for α-Amylase and Its Mutants during an Enzymatic Reaction

【目的】α-淀粉酶是一种重要淀粉水解酶,而Km值是酶反应中重要的参数,尝试建立一种利用α-淀粉酶初级结构定量预测米氏常数Km值的有效模型。【方法】通过神经网络模型,利用535种氨基酸属性定量预测α-淀粉酶Amy7C及其52个突变体反应的Km值,其中33个酶用于模型训练,其余的用于模型验证。首先用双层的20-1前馈反向传播的神经网络进行预测,然后对多层神经网络模型进行筛选。【结果】535种氨基酸属性中有109种属性可以用模型预测,其中动态属性拟合结果较好,4个动态氨基酸属性中有3个属性可以用于模型预测,但拟合结果最好的氨基酸属性分别来自氨基酸理化性质和二级结构。对9种拟合和验证结果最好的氨基酸属性进行7种多层神经网络模型拟合,结果显示增加模型的复杂度并不能提高预测结果的精准度,表明较为简单的模型,如20-1或20-5-1是定量预测建模的首选。【结论】α-淀粉酶酶解反应的米氏常数Km,可以利用某些氨基酸属性通过神经网络模型进行定量预测。为今后利用酶的初级结构定量预测酶反应中各参数最适条件提供思路。

Objective]We attempted to develop models to quantitatively predict the Michaelis-Menten constant Km with information about primary structure ofα-amylase,which is a crucial enzyme forα-1-4 glucosidic linkages hydrolysis in starch,while Km is a very important parame-ter in enzymatic reactions.[Methods]By means of neural network,535 properties of amino acids were used to quantitatively predict Km value ofα-amylase Amy7C and its 52 mutants,which were divided into two datasets,3 3 used for model training and the rest for model validation.The training and validation were conducted firstly by means of two-layer （20-1）feedforward back-propagation neural network,and then by multi-layer neural network models.[Results]Among 535 screened properties of amino acids,109 properties can work as predictor and the dynamic properties give better results with 3 converged out of 4 in 20-1 neural network model.Howev-er,the best predicted results came from the amino acid properties with physicochemical proper-ty and second structure,of which nine predictors were conducted by seven multi-layer neural network models.The results showed that the increase in complexity of predictive models did not give too much improvement,indicating that the simpler 20-1 and 20-5-1 models should be the first choice.[Conclusion]The Michaelis-Menten constant Km ofα-amylase can be quantita-tively predicted by some amino acid properties through neural network,which paves the way for quantitatively predicting parameters in enzymat-ic reactions according to the information of pri-mary structure of enzyme.