Probabilistic searching of databases compared to binary CODIS eligibility: An improved methodology for maximizing...

A DNA reference profile of a person of interest may not always be available for comparison to a questioned profile generated from an item submitted for examination. Reasons may include that no suspect(s) has been developed yet in the investigation. In these situations, searching the questioned profile against a DNA database may generate an investigative lead for further consideration. In the United States, the Federal Bureau of Investigation maintains the Combined DNA Index System, which allows for searching in cases where no suspect(s) has been identified. The highest level of CODIS contains over 20 million profiles in its database; however, it also has the most stringent criteria for searching. These criteria state that a DNA profile must have at least 8 of the original 13 core loci and a Moderate Match Estimate (MME) of 1 in 10 million or smaller. Many mixtures fail this MME threshold, limiting chances for generating investigative leads. This paper examined a probabilistic search strategy in an effort to improve database searches. DBLR™ is a software that can assign likelihood ratios (LRs) to profiles in a stored database of STRmix™ deconvolutions. DBLR™ can store databases larger than the current CODIS database and perform a probabilistic search of each database profile using all of the information in the questioned mixture. In this study, a total of 155 DNA mixtures derived from 81 unrelated individuals from eight laboratories were tested. These mixtures were made up of 2-4 contributors and varied in terms of mixture proportions, input DNA, PCR cycles and STR kits. Each of the 155 mixtures were previously deconvoluted in STRmix™ using the STRmix™ parameters of each lab to yield 765 STRmix™ deconvolutions with a total of 1220 slots available for searching. The deconvolution results were analyzed using a CODIS MME proxy tool to determine how many of the 1220 slots pass the 1 in 10,000,000 MME threshold and therefore would be eligible for upload to NDIS. The same deconvolutions were searched using DBLR™ against a database of 25,000,081 potential donors with an LR threshold of 10 million. The results show that 37.7% of deconvolutions pass the MME threshold and would be eligible for upload to NDIS at CODIS. DBLR™ returned LRs above the parallel threshold of LR= 10,000,000 for 83.4% of ground truth donors while returning 4 non-donor only "hits".