This document describes some extra features not present in the original version of the applets and not described in the original publication. The extensions are enabled by appending ?ext to the page URLs. Once applied, all intra-site links are updated to include it, and it can and should be saved in any bookmarks. It is included in the links on this page.
When the extensions are not enabled, the applets behave exactly like the original version except for the short paragraph about the extensions.
Shift concentration range (the “decade shift”)
This setting (labelled “Shift concentration range” on the simulation pages) shifts the ranges of all sliders and graph axes up or down by whole orders of magnitude — powers of ten, where one “decade” is a factor of 10. The default value 0 spans roughly nanomolar to millimolar (10−9 to 10−1 mol l−1); a positive value lets you explore higher-concentration, weaker-affinity regimes, and a negative value reaches lower concentrations (picomolar and below) for very tight binders. Possible values are between −9 and +4 inclusive. Because the sliders and the axes shift together, the curves themselves do not move — only the numbers on the axes (and any plotted data points) change.
Precise iterative search
A calculation method similar to the iterative search, this attempts to converge to a least squares fit without being restricted to the discrete values of the sliders and thus gives much more precise results. In consequence, if the initial values are far from a good fit, arriving to a least squares solution requires a large number of iterations. The algorithm is implemented using recursion, and it fails if it fills the stack before converging. Therefore, it is recommended to use one of the other calculation methods first and then run this. (I might fix this issue in the future.)
Once a least squares fit is found, the values of the optimised parameters (but not the slider positions) are adjusted accordingly. The result is displayed in a pop-up dialog. If there are more data points than free parameters, estimates of standard errors and uncertainties with a 95% level of confidence are calculated for each free parameter. The estimates are reasonably accurate only if they are relatively small or, in other words, when relative uncertainties are much less than 100%. It is impossible to give universally applicable guidelines on how small they should be, but generally if relative uncertainties are greater than 100%, the results are probably not worth reporting.
Logarithmic residuals
In the original version, least squares fit always calculated residuals by taking the difference between theoretical and experimental (data point) values. With this option enabled, the differences of their logarithms are calculated instead. This usually produces a visually better fit into data points when using a logarithmic (absolute) vertical scale. Whether this is valid to do should be taken into consideration, however.
Other small fixes
There were some inconsistent axis limits that were left in the original version by mistake. Specifically, the lower limit of the logarithmic vertical axis in competing ligands and competing receptors simulations is 10−10 instead of 10−12. This is fixed when extensions are enabled.