Testing likelihood ratios produced from complex DNA profiles

Area of Science:

• Forensic Science
• Statistical Genetics
• Computational Biology

Background:

• Assessing the performance of DNA profile analysis models is crucial for reliable forensic evidence interpretation.
• Current validation methods include simulations, large-scale studies, and theoretical alignment.
• The likelihood ratio (LR) is a key metric in evaluating DNA evidence.

Purpose of the Study:

• To investigate diagnostics for assessing DNA analysis model performance using "Hd true" tests.
• To focus on the proportion of non-contributor comparisons yielding a high LR (p) and the average LR for true contributor comparisons.
• To verify theoretical predictions regarding these diagnostics.

Main Methods:

• Utilized "Hd true" tests to evaluate model performance.
• Focused on two key diagnostics: p (proportion of non-contributor LRs >= LRPOI) and average LR for true contributor tests.
• Applied a continuous interpretation model to nine DNA profiles of varying quality and complexity.

Main Results:

• Theoretical predictions suggest p should be <= 1/LRPOI, making it of limited diagnostic value.
• The average LR for "Hd true" tests is a more robust diagnostic, ideally near 1.
• The study verified these theoretical expectations using real DNA profile data.

Conclusions:

• The average LR for "Hd true" tests is a superior diagnostic for DNA model performance compared to p.
• Continuous interpretation models can be effectively validated against theoretical predictions.
• Rigorous testing using established diagnostics ensures the reliability of DNA profile evidence analysis.