Täljedal I B, Wold S
Biochem J. 1970 Sep;119(2):139-43. doi: 10.1042/bj1190139.
Some analytical functions were tested for goodness of fit to standard curves in radio-immunoassays of insulin (Hales & Randle, 1963). Letting a, b and c be parameters, and y the radioactivity measured at an insulin concentration of x, excellent fit was obtained with the function 1/(y+c)=ax+b. This function was associated with a smaller variance around y and a greater number of sign changes for the residuals (calculated radioactivity minus observed radioactivity) than any of the following: 1/y=ax(c)+b; y=ae(bx)+c; y=a+bx+cx(2)+dx(3). The mean number of sign changes of the residuals in 50 separate assays of mouse insulin was within the 95% confidence interval expected for a correct model. The function that was found to be applicable to assays of mouse, human and ox insulins should be useful in routines for computer processing of insulin assay data and may also prove applicable to other radio-immunoassays.
在胰岛素的放射免疫分析中（Hales和Randle，1963年），对一些分析函数进行了拟合标准曲线优度的测试。设a、b和c为参数，y为在胰岛素浓度为x时测得的放射性，则函数1/(y + c)=ax + b获得了极佳的拟合效果。与以下任何函数相比，该函数在y周围的方差更小，残差（计算得到的放射性减去观察到的放射性）的符号变化更多：1/y = ax(c)+b；y = ae(bx)+c；y = a + bx + cx(2)+dx(3)。在对小鼠胰岛素进行的50次单独分析中，残差符号变化的平均数在正确模型预期的95%置信区间内。发现适用于小鼠、人及牛胰岛素分析的该函数，在胰岛素分析数据的计算机处理程序中应会有用，并且可能也适用于其他放射免疫分析。
