The processing offorensic DNA samplesfromextractiontointerpretation can introducenon-biological, process-related stutter and pull-up artifacts that cansignificantlycomplicateDNA profile interpretationbyappearing as biologically significant alleles.This project aims tomodel stutter and pull-up product formation to enable automated identification and removal ofnon-allelic influence from the profile. The results will provide more accurate and reliable meansof artifact identification and will also contribute to understanding the GlobalfilerHuman DNAAmplificationsystem(ThermoFisher Scientific), a relatively new amplification kit thatiscurrently undergoing widespread implementation. Stutter and pull-uparemodeledusingGlobalFiler-amplified human DNA data.Stutterismodeled using nonlinear regression and pull-upismodeled using symbolic regression via genetic programming, a form of artificialintelligence. Both leverage large datasets and computational methods to determine optimalmodels for artifact detection.The resulting stutter modelsused in conjunction with a dynamic,locus, sample specific analytical threshold (LSST)successfully removed96.91% of stutterfromhumanDNA profiles. Pull-up models usinga dynamic, locus, sample specific analyticalthreshold(LSST)successfullyremoved98.1 % ofpull-up peaksin human DNA profiles overallacross all dye relationships.
