Examining performance and likelihood ratios for two likelihood ratio systems using the PROVEDIt dataset.

Area of Science:

• Forensic Science
• Genetics
• Biostatistics

Background:

• A likelihood ratio (LR) system encompasses the entire process of DNA mixture measurement and interpretation.
• Probabilistic genotyping software (PGS) is a critical component within these LR systems.
• Understanding performance variations between different LR systems is essential for accurate forensic analysis.

Purpose of the Study:

• To evaluate and compare the performance of two independently developed LR systems (STRmix v2.6 and EuroForMix v2.1.0).
• To assess the ability of each LR system to discriminate between true contributor and non-contributor scenarios using a known ground truth dataset.
• To demonstrate the value of using publicly available datasets for validating and comparing LR systems.

Main Methods:

• Generated and deconvolved 154 two-person, 147 three-person, and 127 four-person mixture profiles from the GlobalFiler dataset.
• Utilized two fully continuous probabilistic genotyping software programs: STRmix v2.6 and EuroForMix v2.1.0.
• Evaluated LR system performance qualitatively and quantitatively, comparing numeric LR values and verbal classifications.

Main Results:

• Differences in numeric LR values and verbal classifications were observed between the two LR systems.
• The magnitude and potential explanations for significant differences (≥ 3 on the log10 scale) were analyzed.
• Identified instances of LR < 1 for true contributor hypotheses and LR > 1 for non-contributor hypotheses were discussed.

Conclusions:

• Comparing multiple PGS with similar discrimination power can offer valuable insights for forensic analysts.
• This approach can aid in interpreting complex profiles, including low-template DNA and minor contributors.
• Utilizing diverse datasets and multiple software provides a potential additional diagnostic check for LR system outputs.