When DNA Is Not a Gold Standard: Failing to Interpret Mixture Evidence
Total Page：16
File Type：pdf, Size：1020Kb
© natali_mis | AdobeStock clear signals yields an unambiguous genetic type (“geno - When DNA Is Not a type”). Comparing definite genotypes, relative to a random person, yields a reliable match statistic that numerically Gold Standard: Failing to conveys the probative force of DNA evidence. But most crime scene DNA is now a mixture of two or more people, Interpret Mixture Evidence with good data but less certain interpretation. As the NAS report noted, there may be problems with how the DNA was interpreted, such as when there are mixed samples. orensic science connects evidence through Simplistic interpretation of DNA mixture data often shared characteristics. Markings on a bullet can fails to produce an accurate match statistic or give any Fappear to match grooves in the barrel of a gun. answer at all. While the limitations and liabilities of Latent fingerprints left at a crime scene may be similar unscientific DNA mixture interpretation were recog - to ridge patterns on a suspect’s hand. Tracks in the nized early on, 3 only recently has this profound forensic mud may mirror the treads of a shoe or tire. Police failure come to the fore. Crime laboratories in Austin, gather forensic evidence to help build a case, and Texas, and Washington, D.C., have been shuttered in police dramas on television convey the myth of foren - large part because of failed DNA mixture interpretation. 4 sic infallibility through the “CSI” effect. 1 Virginia re-evaluated DNA match statistics for mixture In 2009, the National Academy of Sciences (NAS) evidence in hundreds of cases. 5 Texas is reviewing 24,000 published its seminal report titled Strengthening Forensic criminal cases for flawed interpretation of DNA mixture Science in the United States .2 The NAS report reviewed evidence. 6 The New York State Police (NYSP) has sup - many forensic modalities and questioned their scientific pressed reliable DNA mixture interpretation methods validity. The interpretation of forensic data is often unre - that could expose its crime laboratory’s mistakes in liable. Match statistics are needed to gauge the strength thousands of cases. 7 These numbers extrapolate to hun - of match between items, relative to coincidence. But dreds of thousands of mixture items throughout the forensic statistics are typically absent or incorrect. United States, and the national press has taken notice. 8 Human bias can skew answers by unconsciously select - This failure of forensic DNA interpretation is of ing favorable data, using knowledge about defendant broad concern. Pervasive errors in DNA match statistics characteristics, or by trying to please stakeholders who undermine public trust in science and erode confidence in have a desired criminal justice outcome. government agencies that misuse science to obtain convic - Deoxyribonucleic acid (DNA) evidence seems tions. A failed DNA gold standard portends little hope for immune to such criticism, long serving as a gold standard fledgling forensic fields. Perhaps the greatest loss is true for other forensic disciplines. Abundant DNA from one justice in a free society. Misinterpreting DNA evidence person produces pristine data signals. Interpreting these causes injustice for defendants denied potentially exculpa - B Y M A R K W. P E R L I N , P H . D . , M . D . , P H . D . 50 WWW.NACDL.ORG THE CHAMPION tory evidence, injustice for victims whose genetic analysis of minute biological shorter than the actual allele (e.g., a 10 cases are lost when inculpatory evidence samples. The first technology was poly - allele with 10 repeated words can pro - is unreported, and injustice for innocents merase chain reaction (PCR), which let duce a fragment having only 9 repeated victimized by crime that DNA could have scientists easily make millions of copies words). Such stutter alleles show around prevented. of small DNA quantities at a genetic 5-15 percent of the true allele’s peak This article reviews the history of locus. 9 The second technology was the height and reside adjacent to the allele failed DNA mixture interpretation. It automated DNA sequencer (now called peak. Stutter peaks can be identified and begins in 1985, at the start of the a “genetic analyzer”), which used elec - removed with single source DNA data, genomics revolution, discussing the ori - trophoretic separation and laser detec - but complicate the interpretation of gins of modern DNA testing. Proceeding tion to measure DNA fragment length mixed or low-level DNA. in five-year increments, it outlines the and quantity. 10 Finally, cheap ubiqui - The original STR loci used in genet - missed opportunities and policy failures tous computing enabled automated ic testing had two letters in a repeated that have resulted in the current situa - analysis of genetic data. 11 word. 15 These di-nucleotide repeats were tion. The article offers recommenda - The STR genetic marker was an popular with geneticists because their tions to help overcome long-standing early beneficiary of this technological high genome density placed them near 12 DNA interpretation problems. juxtaposition. PCR amplification of an most genes. However, they gave complex F STR locus produced DNA fragments in stutter patterns with a long trail of frag - A detectable quantities. Separating frag - ments having from 5 to 10 dropped I Biology L ments on a DNA sequencer showed data words. For that reason, forensic identifi - I The human genome contains three peaks, with longer alleles having greater cation (which had to be explained to lay N billion DNA letters, a text written across length. Computer analysis of STR data juries, and only needs a dozen loci) G 23 chromosomes in the nucleic acid could identify and size these peaks to employed tetra-nucleotide repeats hav - T 16 alphabet A, C, G, and T. This textual infor - indicate allele events and would eventu - ing four letters in a repeated word. O mation is used to operate, maintain, ally automate genotype determination. Their simpler stutter patterns usually evolve, and grow human organisms. Part show just one prominent stutter peak. I of the genome’s power is the encoding of Automated computer analysis N 1990: Threshold T this biological operating system. Another could mathematically separate stutter E 17 aspect is the variation between people The original STR genetic tests were peaks from STR locus data. Some R found in noncoding regions that scientists done on DNA samples from a single genetic and forensic practitioners used P can use to trace ancestry, map disease, and source, not mixtures. The locus data this computerized approach, 18 but R distinguish between individuals. had one or two tall peaks, correspon - most technicians were more comfort - E T Scattered throughout the human ding to the one or two parental alleles in able removing stutter visually. genome are genetic locations (loci) that an individual’s genotype. The testing Other random factors affect geno - D have a short DNA word repeated in tan - was done for genetic diagnosis, genome type data. These largely arise from the N dem. These short tandem repeats (STR) mapping, and drug discovery. 13 inherent random variation in PCR copy - A are a rich source of genetic variation. With simple single source data, the ing. Within a copying cycle, some DNA The number of repeated words at a locus interpretation issue was separating the fragments will copy more efficiently M I varies between different people, and true alleles from background noise or than others. Given identical DNA input, X these STR length variants (alleles) can be data artifacts. This separation was this random copying process introduces T used to identify individuals. accomplished by drawing a line that variation in the data output, with each U A cell nucleus has two complete separated tall allele peaks from short PCR experiment producing its own data R genome copies of the 22 human auto - non-allele peaks. A DNA sequencer pattern. This natural variation in DNA E somal chromosomes, one inherited manufacturer advised setting this counting is well known to scientists, and E 19 from each parent. At a particular locus threshold at around 100 relative fluores - it has been mathematically modeled. V on a chromosome, there are two alleles cent units (RFU). There was no statisti - I – maternal and paternal. A person’s cal science involved, just a rule of thumb D 2000: Mixture E pair of alleles at a genetic locus defines to help technicians interpret their allele N the person’s genotype at that chromo - data based on peak height. The Federal Bureau of Investigation C some location. (FBI) helps regulate forensic DNA analysis E An STR locus with many (for exam - 1995: Variation in the United States. The agency’s Scientific ple, 15) allele variants yields very many Working Group on DNA Analysis genotype allele pair possibilities (for With abundant DNA from one per - Methods (SWGDAM) convenes twice a example, 100). Examining multiple inde - son, and clean data signals, thresholds year to discuss policies of interest to the pendent STR loci multiplies those possi - worked well to separate tall allele peaks FBI laboratory. SWGDAM members are bilities, allowing for a trillion trillion (1000 to 2000 RFU) from baseline instru - forensic practitioners, mainly government possible genotypes (24 powers of ten). ment noise (5 to 15 RFU). However, other employees of crime laboratories or police Since there are fewer than 10 billion peo - data artifacts produced peaks over base - organizations. They are not experts in ple alive today (10 powers of ten), there line, or subtler peak patterns. modern statistical computing and its are far more STR genotypes than people, PCR stutter is an error in the DNA application to interpreting DNA data. making DNA useful for identification. copying mechanism. 14 When the poly - The FBI had developed a popula - merase enzyme copies a DNA region of tion statistics computer program 1985: Revolution STR text, it can lose its place and skip (Popstats) for calculating DNA match over one of the short repeated words.