Parametrically Constrained Spectrum Analysis Fiting Process

The parameters of this dialog define and control an analysis run to find the set of solutes that best fits experimental data.

Each analysis run proceeds over a defined set of curves in s and f/f0 space. The single analysis pass produces the model whose associated simulation differs the least from the experimental data, as determined by its RMSD value. Time-invariant and/or radially-invariant noise(s) may also be computed. The curves in a set cover a specified s and f/f0 range and vary according to a specified increment or count. Each model has a specified number of input solute points along its curve.

After an initial pass in which all the specified model curves are evaluated, the results are sorted by fitness (least RMSD). The best model then becomes the starting point for a second pass that utilizes Levenberg-Marquardt to refine the model. The result of that second pass is inserted as the final best model.

For a straight line type, a specified variation count is given. Lines cover the full s range, with end points along the f/f0 range varying the specified number of times. Each implied starting point connects with each of the possible end points. So the square of the implied number of f/f0 end points is the number of models analyzed. If the type is Horizontal Line, the f/f0 value is the same for start and end points, so the number of models is the variations count itself (not its square).

For sigmoid types, a variation count is directly specified. The curves cover the full s and f/f0 ranges, with “par1” and “par2” values each varying the specified number of times, yielding a number of test models equal to the square of that count. The par1 value changes logarithmically from 0.001 to 0.5; and par2 changes linearly from 0.0 to 1.0.