Prediction of enzyme kinetic parameters based on statistical learning.
Prediction of enzyme kinetic parameters based on statistical learning.
复制标题
基于统计学习的酶动力学参数预测。
DOI：
--
复制
发表时间：
2006
期刊：
影响因子：
--
通讯作者：
E. Klipp
中科院分区：
文献类型：
--
作者：
S. Borger;Wolfram Liebermeister;E. Klipp
酶动力学参数的值是生化系统动力学建模的关键条件。然而，对于大多数动力学参数，甚至没有一个数量级是已知的，所以从实验数据的模型参数的估计仍然是系统生物学的主要任务。我们提出了一种统计方法来推断跨物种和酶的动力学参数的值，利用参数值已被测量在各种条件下，现在存储在数据库中。我们通过统计回归模型拟合数据，其中底物、组合酶-底物和组合生物体-底物对对数参数值具有线性影响。因此，我们得到未知酶参数的预测和误差范围。我们将我们的方法应用于BRENDA数据库中的十进制对数Michaelis-Menten常数，并使用留一交叉验证确认结果，其中我们一次屏蔽一个值并从剩余数据中预测它。对于一组8种代谢物，我们得到预测值与真实值的偏差的标准预测误差为1.01，而实验值的标准偏差为1.16。该方法适用于其他类型的动力学参数，许多实验数据是可用的。
Values of enzyme kinetic parameters are a key requisite for the kinetic modelling of biochemical systems. For most kinetic parameters, however, not even an order of magnitude is known, so the estimation of model parameters from experimental data remains a major task in systems biology. We propose a statistical approach to infer values for kinetic parameters across species and enzymes making use of parameter values that have been measured under various conditions and that are nowadays stored in databases. We fit the data by a statistical regression model in which the substrate, the combination enzyme-substrate and the combination organism-substrate have a linear effect on the logarithmic parameter value. As a result, we obtain predictions and error ranges for unknown enzyme parameters. We apply our method to decadic logarithmic Michaelis-Menten constants from the BRENDA database and confirm the results with leave-one-out crossvalidation, in which we mask one value at a time and predict it from the remaining data. For a set of 8 metabolites we obtain a standard prediction error of 1.01 for the deviation of the predicted values from the true values, while the standard deviation of the experimental values is 1.16. The method is applicable to other types of kinetic parameters for which many experimental data are available.