[ Up ] [ references ] [ troubleshooting and questions ] [ Concepts for Getting Started ] [ sedphat help ] [ download ]
Troubleshooting and Frequent Questions
--> Curves are far from the data, although binding constants, molar masses, and/or s-values are reasonably initialized:
Check that the concentration starting guesses are appropriate. Note that interacting models are in micromolar units, and some of the species models in signal units. See the particular selected model for details on the units.
--> Cannot load data:
* Make sure that no 'white space character' is at any place of the directory name. Don't use names like 'c:\XLA DATA\...' or 'C:\My Documents', and don't use the Desktop.
* After moving files to a different computer, you need to copy both the xp-files and the raw data. Further, you need to restore the precise directory structure, because the xp-files contain an entry with the original path, which is where Sedphat will expect the raw data files to be located. If you cannot restore the same path, you need to edit the xp file (see experiments for details).
--> Cannot reproduce fit:
* Make sure that no 'white space character' is at any place of the directory name of either the data, xp-file, or sedphat file. Don't use names like 'c:\XLA DATA\...' or 'C:\My Documents', and don't use the Desktop.
* After moving files to a different computer, make sure you have all files - the xp-files, the sedphat-files and the sedphat_PAR files in the correct locations.
* After moving files, you need to restore the precise directory structure, because the xp-files contain an entry with the original path, which is where Sedphat will expect the raw data files to be located. If you cannot restore the same path, you need to edit the xp file (see configuration for details).
* the interval of minimum and maximum meniscus position do not bracket the meniscus position. This will produce an OK 'Run', but the constraint of minimum and maximum meniscus will be enforced during the 'Fit', moving the meniscus away from its previous position. Same can happen with the bottom. If meniscus or bottom are floating parameters, uncheck them to keep them fixed. Then Go to experimental parameter box, check the meniscus and/or bottom again to be floated, close the experiment box, and enter better minimum and maximum values when prompted. By toggling the floating option for the meniscus, you can force the prompt for minimum or maximum values.
--> Results from Sedfit and Sedphat are inconsistent
* Sedphat calculates in paramters under standard conditions (s20,w), whereas Sedfit is in experimental conditions.
--> When modeling multiple sedimentation equilibrium profiles taken from the same cell at different rotor speeds, can I constrain the baseline to be the same?
Yes, this can be done if the profiles are loaded as a single multi-speed equilibrium experiment.
--> How can I calculate errors for the parameters?
Errors can be calculated in three ways:
1) Use the F-statistics calculator of Sedfit just for calculating the critical increase of the chi-square of a fit, based on the number of data points you have. With this in hand, go back to Sedphat and calculate the best-fit. Fix the parameter of interest to a value slightly different from the best-fit value, while floating all others, and observe the increase in the chisquare of the fit that this constraint causes. (In cases of complicated error surfaces, you may find an even better fit than the initial one - in this case you need to start over from the new best fit.) Do that in a series of fit, walking your parameter of interest away, until you hit the critical value that on a given confidence level is considered statistically significant worse. This procedure is essentially like the one in Sedfit, described here.
2) In Sedphat, if you use Marquardt-Levenberg optimization, the covariance matrix is generated. Go to the statistics function and pull up the covariance matrix. It has information on the error of the parameters, estimated from the local curvature of the error surface at the best-fit minimum.
3) Another possibility in Sedphat is to do a Monte-Carlo analysis. This is also described in
The errors from the covariance matrix (method 2) are usually pretty good, although frequently slightly underestimating the true errors. The errors from Monte-Carlo are rigorous in theory, but in practice suffer from the fact that the fitting procedure must be automated, which for many models just doesn't work well. Therefore, I usually prefer the method 1 - although manually laborious, it is rigorous, and gives you the best feel for how the parameters are correlated with each other.