The Heat Map graphic allows the visualization and analysis of populations that statistically do not belong to the model. These aberrant populations appear in the heat map with unprecedented clarity and sensitivity. To use this new feature effectively, we need to examine how it works.
Normally when a GemStone Cell Type is created, the Exclusion Probability property is set to 0.05.
The property causes GemStone to exclude some fraction of outlier events from being classified by a Cell Type. If we did not have this option, unwanted events from nearby populations would become part of the Cell Type. For example, if we are modeling bone-marrow B-cells, we do not want myelo-monocytic events to contaminate our B-cell Cell Type. This property ultimately defines the extent of our Cell Type in any number of measurement dimensions.
Let’s suppose for this example that you have a progression similar to that shown below.
This model defines the B-cell progression probabilistically in all dimensions. If we had originally enriched our analysis packet with CD19 SSC B-cells, there could be events that for one reason or another do not belong to this normal progression. The Heat Map graphic is the best way to examine these events. Just click the Heat Map tool button and define the area where you want the graphic using your mouse.
If you want to see all the events in the analysis packet, set your Exclusion Probability property to 0.00. This will cause every event to be classified by your Cell Type. Here is what the heat map looks like for this bone marrow B-cell example.
The X-axis represents our states in the B-cell progression. The Y-axis is the difference between the events and our model. The horizontal gray line is currently set to 0.01. Anything above this line has less than 1 out of 100 chance of being part of our cell-type. The higher the difference (higher on the Y-axis), the more likely the population is statistically different.
We see quite a number of populations in this heat map. The long horizontal bands below the p=0.01 line are presumably due to imperfections in the normal distributions of one or more parameters due to events piling up at the axis origin. The more interesting populations are further up on the Y-axis. If we zoom in on all the populations that are above the 0.01 p-value line, we can clearly see a number of small, outlier populations
What we really want to do here is to see other characteristics of these populations. For example, we might like to see where the cells fall on a CD38 vs. CD20 plot. We can do that very easily. Let’s put a region around one of the populations.
We’ll label this region as “Unknown 1” or “UK1”. We can see that it comprises about 0.12% of our B-cells, or about 0.001% of the total events. To find out where these events are in all the other GemStone plots, we need to use another GemStone feature: Animated Regions.
Animated Regions
You can animate any region in GemStone. Simply right click the region and choose Toggle Region Animation. You can also double-click the region and check the Animate Region property.
What happens when you animate a region? Every graphic in GemStone shows where the region’s events are located by blinking those events with the region’s color. In dot plots, the animated dots are drawn larger than the other dots so it is easy to see the events.
The arrows in the figure above show where the UK1 events are in our example. These events seem to be plasma cells that have low CD45.
You can also use animated regions in reverse, to investigate where known populations fall on the heat map. For example, if we turn off animation for UK1 and turn on animation for R2: Plasma Cells region shown above, we can see the blinking dots in the heat map graphic:
In this example, we can see quite a few plasma cells in the heat map graphic, and they are not all in the same cluster on the heat map. One of the parameters that differentiates these cells is CD45. The plasma cells near B3 have higher CD45 than the ones near B1.
The Heat Map is a great exploratory tool. It can be used to find rare, malignant events in the marrow or to find new normal populations. If tumors have subtly different clones, the Heat Map will show them as separate spots. By using the Heat Map with animated regions together, we can understand characteristics in abnormal cells that were never possible before.