This tutorial introduces the use of more than one Cell Type in a model. It continues the model-building process from Model Building 101, 102, and 103.
When we finished Model Building 103, we had developed a fairly detailed model of B-cell maturation. We used a 12-parameter FCS file to test the model, and we were able to observe correlated transitions in marker expression even for parameters that we did not model.
Our model does not yet account for a branch that occurs in B-cell maturation. As the cells mature, they will either express Kappa or Lambda, but not both. We want to capture this with our model to quantify the relative fractions of Kappa and Lambda cells, and to see if other markers have different characteristics for the two subsets.
To do this, we will create a second Cell Type. In general, whenever you need to analyze more than one population or subset, you will need to create a separate Cell Type for each. We will handle the mechanics of doing that here.
We’ll start by opening the model that we have developed.
Click Open Document on the main toolbar. Select “Model103.gs” and click Open.
The model will be read and displayed in its last state.
We’re going to make the first Cell Type for Kappa B-cells. We will select Kappa cells using a Constant profile on the Lambda parameter to identify Lambda-negative cells. The advantage to this approach is that our model will be able to show clonal excess for Kappa cells. Saying that another way, the model will not filter out events that are very positive for Kappa.
Click the Cell Type shrink tool to expose the editors we will be working with.
In the Cell Type Properties panel, change the Cell Type Name to “Kappa”, and edit the Abbreviation to “K”.
Locate the Lambda parameter in the Cell Type widget. It should be the last parameter. Click the Lambda label to activate the parameter.
In the Parameter Profile property panel, choose Constant for the Parameter Profile. Also, enable the Lock Fit checkbox. This tells GemStone not to move this profile, which is very important when identifying “negatives” in this manner.
Position the Control Points for the profile to identify the low-intensity Lambda events as shown.
We will not analyze the data yet. First, let’s create the Lambda Cell Type.
The only difference between the Kappa and Lambda Cell Types will be in the use of the Kappa and Lambda parameters. All of the other parts of the model for these two Cell Types will be the same. That means that we can duplicate the work we’ve done already and the make a few changes in the second Cell Type to model the Lambda cells.
Click the Utility Action button on the Workspace toolbar, and select Duplicate Cell Type from the menu.
GemStone will make a copy of all of the elements in our Kappa Cell Type, and name the copy “CellType2”. It also switches to the tab for CellType2.
Model Building Tip: Duplicate Cell Type
Use the Duplicate Cell Type command to make a copy of an existing Cell Type. The command will copy the model definition, as well as the layout and graphics for the Cell Type.
Click the Cell Type shrink tool to expose the editors for CellType2.
In the Cell Type Properties panel, change the Cell Type Name to “Lambda”, and edit the Abbreviation to “L”.
Now we need to make two more changes. We have to turn off the parameter profile for Lambda in this Cell Type. Then, we need to model the Kappa parameter to select Kappa negative events.
Locate the Lambda parameter in the Lambda Cell Type widget. Click the Lambda label to activate the parameter.
In the Parameter Profile property panel, choose None for the Parameter Profile.
Next, locate the Kappa parameter in the Lambda Cell Type widget and click the Kappa label to activate the parameter.
In the Parameter Profile property panel, choose Constant for the Parameter Profile. Enable the Lock Fit checkbox as we did for the Lambda profile in the Kappa Cell Type. Position the Control Points for the profile to identify the low-intensity Kappa events as shown.
This is a very useful B-cell model at this point. We should save a new copy of it.
From the main toolbar, click Save Document. Type “Model104.gs” for the file name and click Save.
We’re ready to analyze our full Kappa-Lambda B-cell model. Let’s use Auto Analysis to do this. Auto Analysis does many of the steps we have been doing manually: it adjusts the constant selection parameters, it enriches, and it then adjusts the parameter profiles for each of the remaining parameters (except those with the Lock Fit option enabled).
Click the Auto Analysis button on the main toolbar.
Like the Classify Data command, Auto Analysis acts on all of the Cell Types that are defined in the model document. An event can be classified as belonging to one of the cell types or it can be considered unclassified. An event can never be classified into more than one of the cell types at the same time. In the model we have created, the Kappa and Lambda Cell Types are assigned events probabilistically.
Model Building Tip: Stopping Auto Analysis
There may be a point at which the RCS and frequency distribution are "good enough" and the optimization is no longer making real improvement to the fit of the model. You can stop the current estimation or the auto analysis process by clicking the appropriate check box in the status bar.
Let’s take a look at the Parameter Overlay plot for each of our two Cell Types. In the Lambda Cell Type’s overlay, we can see the elevated expression of the Lambda marker starting late in zone III. The Kappa expression for Lambda cells remains low.
Conversely, in the Kappa Cell Type, the Kappa expression increases in zone IV, and Lambda expression remains low throughout.
In this series of tutorials, we have developed a GemStone model from scratch to a point where it is a very detailed analysis of B-cells with Kappa and Lambda differentiation.
We outlined a model-building strategy where we design part of a model, synthesize data with GemStone, increase the complexity of the model, and continue this cycle to a point where we are ready to test the model on real samples. This approach allows us to test our assumptions about what we “know” about the cells we are going to analyze. The use of synthesized data makes it much easier to increase complexity gradually in model development, and as we saw, the synthesized events take on very “real” characteristics as we build the model.
The model-building tips we learned along the way are worth reviewing here. Models should start out by defining the Constant parameters, which we referred to as selection parameters. These are the simplest parameters for us to set up and for GemStone to analyze. Then we set up simple parameters that have the most dynamic transitions over the progression. The dramatic transitions are easier for us to visualize, and we can make confident decisions about where the control point should be positioned. Finally, we modeled some of the more complicated parameters. GemStone helped us with these because of the simple parameters we modeled first.
We created zones for analysis and used the Parameter Overlay plot to view the correlated expression of all of the parameters in our files. And finally, we learned how to handle subpopulations (Kappa and Lambda) with multiple Cell Types.
Use these principals to develop models for the cells you want to analyze, and you should produce robust models that can be used with GemStone for a great variety of applications.