Flow cytometry data analysis focuses on identification of populations in a sample and providing statistical results about those populations. Typically, regions are drawn on univariate or bivariate plots, and then used to create a hierarchy of subsets by “gating” the events. The resulting subsets are analyzed for frequency, intensity, and other common statistics to characterize the cells that were analyzed.
Cytometrists are trained to identify subsets of interest. We learn the characteristics of the subsets and which parameters are used to select them out of the many cell types that our samples contain. The learning curve can be long, and we often find it difficult to explain our plots and strategies with other scientists unfamiliar with cytometry.
GemStone approaches data analysis quite differently by using a Probability State Model (PSM) to identify and quantify subsets. So what is a Probability State Model, and why is it a better approach?
Let’s examine the words that make up the name to get an idea of what a PSM is.
Probability: the statistical likelihood of a particular event occurring. A PSM classifies events using probabilities rather than user-defined gates. The gating approach is subjective, and relatively small errors made in drawing gates are compounded with each subsequent gate in the hierarchy. By using probability, a PSM actually reduces the error with each additional parameter that is added to the model. The likelihood that you are actually identifying the cells you are interested in increases as you build a PSM.
State: a set of conditions that discretely characterize something. A PSM probabilistically classifies events into a set of states, based on the characteristics defined for each parameter. The state for each event can be thought of as an additional parameter, one that identifies the particular set of characteristics that define that state. GemStone uses the state as the common, X-axis for its parameter plots. This allows all parameters to be compared on a common axis, making it possible to correlate and compare all of the parameters in a single plot.
Model: a mathematical representation of a process. A PSM uses fitting routines to classify cells into the most probable states, or to adjust the model to match the data. Using this approach, GemStone can provide objective measurements of how well the model represents the data. There are no hard-edges in a PSM. The modeling process allows populations to overlap based on probabilities.
A PSM allows us to classify events into populations probabilistically based on a model that we define. The model is made up of one or more Cell Types. Each Cell Type defines the characteristics of a different subset that we want to analyze. We use what we know about the markers in our experiments to create a set of Parameter Profiles for each Cell Type subset. A Parameter Profile uses a set of control points to define how the subset transitions over the state index axis.
The state index axis is typically labeled “Progression”, but it does not have to be a progress at all. In fact, simple models can be designed to identify and quantify cell populations where the parameter profiles are simply set for constant intensities.
The PSM approach allows cytometrists to evaluate their analysis more objectively and more scientifically. By using a model, we can measure and report confidence limits. We can evaluate the goodness-of-fit with reduced chi-square (rcs). And we can co-plot our parameters in a single graphic that allows us to examine the coordinated transitions of markers in the cells we analyze.
See also: