Up ] references ] troubleshooting and questions ] [ Concepts for Getting Started ] sedphat help ] download ]

Concepts for Getting Started

This tutorial assumes general familiarity with the experimental setup.  For detailed information, see the experimental protocols on the website of the PBR/DBEPS at NIH.

As in Sedfit, the basic sequence of using Sedphat is 

1) load data

2) select the analysis model

3) enter starting guesses and fit 

4) save and document results, save Sedphat configuration

5) make statistical analysis of confidence

perhaps 6) re-analyze the data with alternative model

A little bit more detailed tutorial for the workflow with Sedphat is available here, but I strongly encourage you to familiarize yourself first with the following terminology and concepts.

Sedphat is a very flexible platform for global modeling. Global modeling is very powerful but requires some more thought about the different data sets, how to assemble them, and about the fitting parameters.  

New in version 1.9b:  Two tools for enabling the storage and reproduction of results have been implemented:

* saving global model parameters into files

* saving the complete configuration of Sedphat

Note also that there are a some possibilities to customize startup defaults and model defaults.

Following are basic introductions to the general concepts.  For more detailed information, see

partial specific volumes
tutorial for getting started
sedimentation equilibrium
sedimentation velocity for reactive systems
sedimentation isotherm analysis
isothermal titration calorimetry
surface binding isotherms


Data Channels ('Experiments')

The major principle of organization is to use different data channels, or 'experiments'.  They can be data from analytical ultracentrifugation or dynamic light scattering, and be of different type: 

* a single sedimentation equilibrium profile

* multiple sedimentation equilibrium profiles from the same cell at different rotor speeds ('multi-speed equilibrium') 

* a set of sedimentation velocity data

* an autocorrelation function from dynamic light scattering

* an isotherm of weight-average sedimentation coefficients or molar masses

Please note:  The term 'data channel' is more accurate than the term 'experiment' which is used in Sedphat synonymous to 'data channel'.  The reason is that multiple data sets can be obtained from a single sedimentation experiment, because the ultracentrifugal cell can be scanned at different wavelengths or with different data acquisition systems.  The data at each wavelength would be loaded in a separate data channel, i.e. a separate 'experiment' in Sedphat, although they are acquired from the same physical system.  Similarly, in dynamic light scattering autocorrelation functions from could be collected from the same sample but at different angles - they are physically the same experiment, but would be considered different data channels or 'experiments' in Sedphat.

Each data channel in Sedphat requires information on the experimental parameters, such as buffer density, viscosity, extinction coefficient, meniscus, bottom, fitting limits, expected baseline parameters, etc.  The complete set of this information can be stored in an experiment file (*.xp), and later retrieved, thus simplifying the process of reloading and reproducing the analysis, or re-assembling data channels in different combinations.

Typically, several 'experiments' (*.xp files) are then loaded into different data channels and combined in for a global analysis.  (Sedphat also functions well for single analysis.)

More details on how to deal efficiently with different experiments, load, save, and remove them, and some examples can be found here.

Obviously, many of the house-keeping operations, like copying data or setting local parameters, etc., require specific user input to direct them to the desired experiment.


Local Parameters, Global Parameters, and Shared Local Parameters

Some of the experiments come with unknown parameters.  For example for a sedimentation velocity experiment, the meniscus position may not be known and best treated as a floating fitting parameter.  Other examples for parameters that are specific to an individual experiment may be the baseline (or radial-baseline profile), or the loading concentration.  Such parameters are referred to as Local Parameters, because they describe aspects of the particular experiment (they are 'local' to that experiment), as opposed to the Global Parameters, which describe overall aspects of the model independent of any particular experiment.  Examples of Global parameters are binding constants, sedimentation coefficients, molar mass values, etc.

It is very useful to introduce a third kind of parameters - Shared Local Parameters.  These describe aspects of the experiments, like local parameters, but they are shared by more than one experiment.  For example, if two data channels consist of sedimentation data from the same cell, but acquired at different wavelengths, it is certain that the meniscus and bottom position of the solution column are the same.  Treating the meniscus and bottom as a local parameter constrained to be the same in each data channel, i.e. a shared local parameter, is very helpful to tighten up the model.  Another example of shared local parameters might be the loading concentration - since the analysis is using molar concentrations, it has to be the same if the data in two channels are from the same cell but at different wavelength.  (This permits to keep the extinction coefficient in one of the data channels a floating parameter.) 

More details and examples of parameters and their combinations can be found here.

Note that part of the global model is the partial-specific volume of the macromolecules under standard conditions.  This is because the partial-specific volume may be concentration dependent, or dependent on buffer conditions.  More no this issue is described under partial-specific volumes.


Selecting the Model

There are a variety of models, which can be grouped in three categories:  Those for non-interacting species, self-association, and hetero-association.  Note that not all models are implemented yet for all data types.  

The non-interacting models come either as continuous distributions, with discrete species, or in hybrid models combining both continuous distributions and discrete species.  

Global analysis also raises the possibility for the models to be distinguished as to whether they require the local concentrations of the species to be the same in all data channels, or if the concentrations are local to each experiment.  The latter could be the case, for example, when different physical experiments or different techniques are used, whereas the former could apply if we're looking at different wavelength data acquisitions of the same physical experimental configuration.  

Sedimentation equilibrium data have the additional possibility to be analyzed with mass conservation constraints, or without.  They are fully capable of global multi-wavelength or multi-signal analysis.

Model parameters can be saved and restored, and their default parameters can be changed.  More on models can be found here.


Local and Global Runs and Fits 

After you select the model, you can apply it globally to all loaded data channels, or you may want to look just at a single experiment.  Accordingly, the Run and the Fit commands split up into local and global versions.  

For example, when you decide to include an additional data set into the analysis, you may want to first get the local parameters of this experiment into a good range. This could be achieved without moving your global parameters if you constrain the global parameters, float the local parameters, and then do a local fit.  This will adjust the local parameters to the best possible values that is most consistent with your prior estimates of the global parameters.  After that, you may want to perform the global fit with floating the local and global parameters in order to get the best overall and utilize the information of the additional data set.  This procedure will avoid having bad local parameter values for the new data move your good global estimates too far away from the optimal values.  

As in Sedfit, Sedphat distinguishes the 'Run' and 'Fit' commands:  'Run' will simply execute a model and not optimize any of the parameters, except some of the linear local parameters, like radial-dependent baseline profiles.  This is in order to allow you to test your initial estimates, and manually set them to reasonable starting estimates.  The 'Fit' command then does the non-linear regression.  This is important because the global models can be complex, and the error surface may contain many local minima.  As always in non-linear regression, having good starting estimates is essential.  

For fitting, both the Nelder-Mead simplex algorithm and the Marquardt-Levenberg algorithms are implemented (and can be selected in the options).  This is useful because the error surface of the global model may be complicated, and non-linear regression non-trivial.  Having two algorithms helps to ensure that a global minimum has been found.  Usually, I use both optimizers sequentially to make sure they converge to the same optimum and no further improvement is possible.

See also the reference site for the Fit command.

For the statistical analysis of the results, both Monte-Carlo analysis and covariance matrix is available.  Dependent on the data and the models, both have their problems and potential pitfalls (see the statistics menu).


Saving the Configuration

The term configuration stands for the combination of a set of experiments, the model, the model parameters, and the links between the experiments and shared parameters.  It is useful to save the configuration of Sedphat in order to document the results, and in order to enable the precise reproduction of the current state of the analysis.  Once a configuration has been saved to the disk, it's name appears in the Sedphat window title bar, and it can be updated as the analysis proceeds.

A configuration on the disk is a ANSI text file, which can be edited with any text editor.  For more details see Configurations.


Documentation of the Fit

There are no sophisticated printing or plotting capabilities in Sedphat.  This is because everybody usually has his own favorite layout and plotting program.  Like in Sedfit, therefore, publication-quality graphs should be made by copying the data, fit, residuals, and/or distributions and pasting them into spreadsheets in your favorite plotting software.  (I have made templates in Origin, but probably SigmaPlot or Excell can work nicely as well.)

An important tool to store results is to save the Sedphat configuration (see above). 

For documenting the detailed fitting results, Sedphat displays all relevant information as text output in the Sedphat window after fitting.  I recommend maximizing the Sedphat window, and using the 'Prnt Scrn' button to put a screendump bitmap in the clipboard.  Then, paste it into any word processing or other software as a picture (I use Powerpoint, shrinking the bitmap to fit into a slide).  Make additional notes there and save the whole document.  In this way, you can get all the details on different alternative models saved. 

There are two tools for visualizing separately the contributions of each species to the total fit. They are controlled in the Lamm equations option box in the Options menu.  They permit to temporarily manipulate extinction coefficients of species (and the profiles can then be extracted with the "copy fit data" function), or to save the molar distributions of each species to the harddrive. 


More specific information on the concepts and organization of Sedphat is available for the topics:

partial specific volumes
tutorial for getting started
sedimentation equilibrium
sedimentation velocity for reactive systems
sedimentation isotherm analysis
isothermal titration calorimetry
surface binding isotherms