You have to choose a regression model (the order of the polynome) for each individual test. This allows for a maximum flexibility of the regression model. You may choose the order of the polynome that is being used to fit the regression. If you choose '1' you will get a straight line, if you choose '7' you will get an interpolation (i.e. the regression line will pass through all data points). Polynomes, especially at a higher order, have a tendency to produce undesirable oscillations. Choose an order that gives a good empirical fit.

Switch from the 'Data Entry' sheet to 'Col 1' (use the tabs on the lower part of the window). The 'Col 1' sheet will contain all regression parameters (such as effective concentrations) and the corresponding graph for the first column of data (A1 to H1). Check the correlation coefficient (R2) and the empirical fit of the data by looking at the regression line in the graph.

Adjust the order of the polynome until you get satisfactory results. Always start at 7 and make your way down. Often 7 will give very good correlation coefficients (1.00) but may result in undesirable oscillations of the regression line.



Switch to 'Col 2' etc. and adjust the order of the polynome for the second column.

EXAMPLE:

The following data are a little tricky. Above IC50 the data points seem to follow a perfect sigmoid curve. Around the first and second data points, however, an outlier significantly disturbes the regression. If the order of the polynome is set to 7 the data result in a perfect correlation (as the predicted regression goes through all data points). The polynome produces undesirable oscillations that render the regression useless. To obtain a good empirical fit it is therefore crucial do adapt the order of the polynome.



After adjusting the order of the polynome to 4 we get a much better empirical fit of the regression, although R2 is now only 0.97. Around IC50/IC90 the fit of the regression is excellent. In this example the lower order of the polynome therefore results in much better estimation of the principal parameters of regression (IC50, IC90).