Investigation into the performance of different models for predicting stutter.

Area of Science:

• Forensic Science
• Statistical Genetics
• Population Genetics

Background:

• The stutter ratio (SR) is a critical parameter in forensic DNA analysis, influencing profile interpretation.
• Accurate modeling of SR is essential for robust statistical inference in DNA mixture analysis.
• Existing models may not fully capture the complex behavior of SR observed in forensic datasets.

Purpose of the Study:

• To evaluate and compare five statistical models for describing the behavior of the stutter ratio (SR) in DNA profiles.
• To identify the most suitable model for accurately representing SR in forensic DNA interpretation.
• To assess the applicability of these models to different DNA profiling systems.

Main Methods:

• Examined two log-normal models, two gamma models, and a two-component normal mixture model.
• Applied each model to single-source DNA profiles from Identifiler™, NGM SElect™, and PowerPlex® 21 systems.
• Evaluated model performance using log-likelihood calculations and Akaike Information Criterion (AIC) differences.

Main Results:

• The two-component normal mixture model consistently outperformed all other models evaluated.
• This model utilizes two distributions with the same mean but different variances to describe SR at each locus.
• The superior performance was observed across different DNA profiling datasets, demonstrating broad applicability.

Conclusions:

• The two-component normal mixture model provides a statistically robust and intuitively appealing approach for modeling stutter ratios.
• This model can enhance the accuracy of DNA profile interpretation in forensic science.
• Implementation in expert systems with continuous DNA interpretation methods is feasible and recommended.