ModFit LT uses some very sophisticated logic when it attempts to automatically analyze your data. The system has been designed to work with a great variety of data samples, samples that vary in almost an unlimited number of ways. When your data samples fit into the assumptions built into the automatic analysis functions, the program works well in automatically modeling them. Conversely, as you might expect, the automatic systems do not work as well when your samples differ from what it has been designed to analyze.
To better understand the strengths and limitations of automatic analysis, it may be helpful to examine some of the assumptions built into the system.
Peak Finder
When you open a data file with the program, one of the first things the software attempts to do is locate "peaks" in the sample. This step is the foundation of automatic analysis: if it finds the important peaks in the sample, the remaining steps in auto-analysis have a good chance at succeeding. If it cannot detect the important peaks, it will more than likely have difficulty or may fail to automatically analyze the sample.
While it is relatively easy for our eyes to see peaks in a histogram, the computer algorithms that perform this function are anything but simple. The process ModFit LT uses works well with the large library of data files we use to test the system. Here is how it works.
The program performs a statistical analysis of the histogram to identify potential peaks. Then the peak finder begins to filter out peaks that do not meet certain criteria. There are filters for high and low %CV, relative peak height, and how well the peak can be fitted with a Gaussian in a quick analysis. These filters can be viewed in the Peak Finder Settings dialog.
If one or more of the filtering criterion filters out an important peak, the system will likely fail in a later step. For this reason, you can adjust some of the filters used by the peak finder so that the system better matches the data samples analyzed in your laboratory.
A very important note on the peak finder settings: adjusting one or more settings to work with a single sample is a bad idea. The reason is simple: without looking at numerous samples to test your new setting, you may find you have fixed the peak finder for one exceptional sample, and broken it for the typical sample. Weigh your adjustments carefully. Do not make adjustments to accommodate a single sample; use manual analysis to handle the exception.
Ploidy Detection
The peak finder identifies the peaks it finds by displaying a small, black triangle under the peaks on the histogram display. The next step in automatic analysis is to look at the pattern presented by those peaks and attempt to match the pattern to a type of histogram the program knows about. This occurs when you click the Auto button on the toolbar, or choose Automatic Analysis from the Analysis menu.
So just what is an "important" peak?
The most important peaks the program needs to find are peaks for internal standards, apoptosis, and G0G1 peaks for each cycling population. If the program finds these peaks, it generally succeeds in auto-analysis.
You need to tell the program how many standard peaks your sample contains. This is done in the Auto Analysis Settings dialog box. The program assumes that standard peaks, if any, are the first peaks, followed by an apoptosis peak, if selected, and the remaining peaks are cycling population peaks.
If you have correctly informed the program as to how many standard peaks the sample contains and whether or not it has an apoptosis peak, and the peaks found by the peak finder are the important peaks, the ploidy detection step is off to a good start.
The remaining step is a pattern-matching step, which ignores standard and apoptosis peaks. Here, the program analyzes relative sizes and positions of the peaks remaining, looking for cell cycles. ModFit LT is designed to handle any number of cell cycles in a histogram, though it is rare to find samples with more than 4 cell cycles.
In general, the more complicated the histogram, the more difficult it is to classify. You may need to manually analyze very complex samples.
Once the sample is classified, a model is created to match the pattern. Then ranges are assigned to important peaks, and the non-linear least-squares analysis of the sample is performed.
Debris and Sample Quality
As you might expect, the success of automatic analysis is closely linked to the quality of the data you want to analyze. Samples with lots of debris or aggregation detract from the program's ability to find peaks or determine the right model to use. Efforts to gate out debris and aggregation sometimes work, but they are usually going to work against the program's built-in functions designed to compensate for debris and aggregates. The better solution is to perfect the sample preparation and acquisition techniques you use. Starting off with a cleaner sample greatly simplifies the peak and ploidy detection operations, and helps ensure that the peaks seen by you and the program are real peaks.
Follow the suggestions of experts to prepare and analyze your samples. Establish standards that must be met before a sample can be considered for analysis. The DNA Consensus Conference (Cytometry. Vol. 14. Number 5, 1993) made many specific recommendations on DNA analysis, many of which relate to sample quality. Your standards should be at least as restrictive as those presented in that report.
Be aware of obvious things, as well. If you accidentally try to analyze forward angle light scatter instead of the DNA parameter, you will end up with nonsensical information when you perform automatic analysis.
Your Responsibilities
Based on this discussion, it is probably obvious that there are numerous ways automatic analysis of your samples can fail. The problem is that the program itself doesn't know when it has failed; only you can make that determination.
You should view automatic analysis with scientific skepticism. You must review the information on the report to ensure all the model matches your understanding of the histogram. You should always review the results of an automatic analysis to ensure that the program has identified the standard peaks, apoptosis peaks, and G0G1 peaks of the cycling populations. In addition, you should verify that the ploidy classification assigned by the program concurs with your knowledge of the sample.