Hopkins J C, Leipold R J
DuPont Merck Pharmaceutical Company, Cardiovascular Diseases Research, Wilmington, DE 19880, USA.
J Theor Biol. 1996 Dec 21;183(4):417-27. doi: 10.1006/jtbi.1996.0232.
Mechanism-based mathematical models describe systems in terms of identifiable physical processes, and the parameters are assumed to have fundamental physical significance. Ideally, the parameter values are measured independent of the system being modeled, but these values are often adjusted to give the best fit of model predictions to experimental data. A systematic investigation of the effects of such parameter adjustment was conducted by developing a model system comprising a known reaction mechanism and known rate constants. Simulations of experiments were run, and then attempts were made to model the system under a variety of problematic, but realistic, conditions. (1) When one rate constant was seriously in error, adjustment of a different rate constant gave the greatest improvement in the model fit. (2) When a contaminant was present in the experiment, the effects could be hidden by the adjustment of the rate constants. (3) When an incorrect reaction mechanism was assumed, the error could be hidden by parameter adjustment if the concentrations of only one of the reacting species were considered or if an unweighted fit was used for the optimization. (4) Parameter values adjusted for one set of experimental conditions gave a poorer fit than did the unadjusted parameter values when attempting to model a new set of experimental condition (addition of an inhibitor). These results show the potential dangers of adjusting parameter values and the importance of measuring as many variables as possible in a complex system.
基于机制的数学模型根据可识别的物理过程来描述系统，并且假设参数具有基本的物理意义。理想情况下，参数值的测量独立于所建模的系统，但这些值通常会被调整，以使模型预测与实验数据达到最佳拟合。通过开发一个包含已知反应机制和已知速率常数的模型系统，对这种参数调整的影响进行了系统研究。进行了实验模拟，然后尝试在各种有问题但现实的条件下对系统进行建模。（1）当一个速率常数严重错误时，调整另一个速率常数能使模型拟合得到最大改善。（2）当实验中存在污染物时，速率常数的调整可能会掩盖其影响。（3）当假设了错误的反应机制时，如果只考虑一种反应物的浓度或者在优化时使用未加权拟合，参数调整可能会掩盖误差。（4）针对一组实验条件调整后的参数值，在尝试对新的一组实验条件（添加抑制剂）进行建模时，其拟合效果比未调整的参数值更差。这些结果表明了调整参数值的潜在危险以及在复杂系统中测量尽可能多变量的重要性。
