A sensitivity analysis to determine the robustness of STRmix™ with respect to laboratory calibration. (July 2018)

• Record Type: Journal Article
• Authors: Kelly, Hannah; Bright, Jo-Anne; Kruijver, Maarten; Cooper, Stuart; Taylor, Duncan; Duke, Kyle; Strong, Melissa; Beamer, Vickie; Buettner, Christina; Buckleton, John
• Is Part Of: Forensic science international. Volume 35(2018)
• Page Start: 113
• Page End: 122
• Publication Date: 2018-07
• Subjects: Interpretation -- Mixtures -- Likelihood ratio -- PCAST report -- Probabilistic genotyping -- Validation; Forensic genetics -- Periodicals; Génétique légale -- Périodiques; Forensic genetics; Electronic journals; Periodicals; 614.1
• DOI: 10.1016/j.fsigen.2018.04.009
• Languages: English
• ISSNs: 1872-4973

Abstract

Highlights: A public dataset, PROVEDIt, is calibrated for STRmix™. A subset of mixtures from this dataset are analysed and any false positives (adventitious matches) and false negatives (exclusions) investigated. The effect of using of parameters developed for one laboratory to interpret mixtures developed at another is investigated. The difference in performance is small. Abstract: STRmix™ uses several laboratory specific parameters to calibrate the stochastic model for peak heights. These are modelled on empirical observations specific to the instruments and protocol used in the analysis. The extent to which these parameters can be borrowed from laboratories with similar technology and protocols without affecting the accuracy of the system is investigated using a sensitivity analysis. Parameters are first calibrated to a publicly available dataset, after which a large number of likelihood ratios are computed for true contributors and non-contributors using both the calibrated parameters and several borrowed parameters. Differences in the LR caused by using different sets of parameter values are found to be negligible.