Constrain tab

Setting constraints is often essential to getting useful results.

Prism lets you constrain each parameter to a constant value, constrain to a range of values, share among data sets (global fit), or define a parameter to be a column constant. This is an important decision, and will influence the results.

You won't always want to fit all the parameters in a model. Instead, you can fix one or more parameters to constant values. For example, if you have normalized a dose-response curve to run from 0 to 100, constrain Top to equal 100 and Bottom to 0.0. Similarly, if you have subtracted a baseline so you know that the exponential decay curve has to plateau at Y=0.0, you can constrain the Bottom parameter to equal 0.0.

Remember that Prism has no "common sense", does not know how you did the experiment, and can't read your mind. Setting constraints is your job.

Constrain to a range of values to prevent Prism from letting parameters take on impossible values. For example, you should constrain rate constants to only have values greater than 0.0, and fractions (say the fraction of binding sites that are high affinity) that have a value between 0.0 and 1.0. Setting this kind of constraint can have three effects:

•	If nonlinear regression finds the best-fit values without ever running into the constraint, then setting the constraint had no effect. you would get exactly the same results without the constraint.

•

If nonlinear regression would get "confused" and set parameters to values that make no sense, then setting a constraint can be very helpful. During the iterations, the curve fitting process may hit a constraint, but then it 'bounces back' and the best-fit value ends up in the allowed zone. Setting the constraint prevented nonlinear regression iterations from going astray, and you can interpret all the results normally. The only way to know this happened is to compare results with and without the constraint, but there is no point in doing this as the results are interpreted the same either way.

•	In some cases, the curve fitting process hits a constraint but isn't able to 'bounce back'. Prism reports that the best-fit value is right at the constraint border, and reports (at the top of the results) that the fit "Hit constraint". This usually means that you set the constraint incorrectly.

If you are fitting a family of curves, rather than just one, you can choose to share some parameters between data sets. For each shared parameter, Prism finds one (global) best-fit value that applies to all the data sets. For each non-shared parameter, the program finds a separate (local) best-fit value for each data set. Global fitting is a very useful tool in two situations:

When you fit a family of curves at once, you can set one of the parameters to be a data set constant. Its value then comes from the column title, which can be different for every data set. This parameter becomes almost a second independent variable. It has a constant value within any one data set, but a different value for each data set. For example, when fitting a family of enzyme progress curves in the presence of various concentrations of inhibitor, the inhibitor concentration can be entered into the column title of the data table. View an example.