Peak Fitting and Peak Attribute Extraction from Capillary Electrophoresis Run Data of Short Tandem Repeat DNA Samples

The Laboratory for Information Technologies at the University of Tennessee is developing an expert system to interpret STR DNA analytic results. An important component of the expert system fits detected peaks with analytical model equations. Information from each fitted peak is subsequently used by expert rules to identify the nature of the peak and determine its allele designation. This report documents the formulation of the peak fitting problem, discusses the various solution approaches, develops the structure of the peak fitter framework, and presents test results used to evaluate its performance.

Using STR DNA data, it was found that most peaks can be fitted well with a simple Gaussian equation. However, less ideal peaks that exhibit asymmetry or have dual peaks require more advanced model equations for an acceptable t. Several modified Gaussian equations are considered and adapted to accommodate these more complicated peaks and obtain satisfactory results. Heuristic criteria have been developed to determine if a current t is acceptable, and if not, which advanced t type is to be attempted next. Statistics that summarize the peak fitter's application to over fifteen thousand peaks of various types are presented, and an analysis shows that almost all peaks can be fitted satisfactorily. Heuristic criteria employed by the peak fitter framework can be tuned to match the characteristics of the DNA data generated by a specific DNA laboratory using standard laboratory protocols and instruments, thereby improving system performance. However, tuning rules are complex, and further research is needed to find an optimal tuning mechanism.