29 July 2008

DNA Evidence Accuracy

DNA evidence is accurate, but nowhere near as accurate as it as sold as being.

In 2001, Arizona crime-laboratory analyst Kathryn Troyer was running tests on the state's DNA database when she stumbled across two felons with remarkably similar genetic profiles.

The men matched at nine of the 13 locations on chromosomes, or loci, commonly used to distinguish people.

The FBI estimated the chances of unrelated people sharing those genetic markers to be 1 in 113 billion. But the mug shots of the two felons suggested that they were not related: One was black, the other white. . . .

Although a person's genetic makeup is unique, his "genetic profile" — a sliver of the full genome — may not be. Siblings often share genetic markers, and unrelated people can share some by coincidence.

No one knows how rare DNA profiles are. The odds presented in court are best estimates from the FBI. . . . Defense attorneys seized on the Arizona discoveries as evidence that genetic profiles match more often than statistics imply and are not unique.

Now, lawyers around the country are asking for searches of their own state databases. Scientists and legal experts want to test the accuracy of statistics using the nearly 6 million profiles in the Combined DNA Index System, or CODIS, the national network that includes state and local databases. . . . in two states . . . searches found nearly 1,000 more pairs that matched at nine or more loci.


Put another way, the data from the news story quoted above suggests that on the order of 1% of people in CODIS have one other unrelated person in the database who would could come up as a match to a sample of their DNA. The risk would be higher for related persons who are both in the same database. Moreover, this is "fundamental risk" which can't be eliminated by better lab technique, and there is some additional risk of laboratory error.

This is still extremely accurate. About 99% of people have a unique DNA profile within the CODIS database, and narrowing a list of suspects to two people, one of whom will often be ruled out as a suspect by incontrovertable alibi evidence or a lack of any connection to the crime, still has incredible evidentiary power. Even when there are a pair of matching suspects, a pure DNA match has a 50% chance of being correct. So, overall, the accuracy of a pure DNA match to a suspect is about 99.5%. DNA evidence remains the gold standard of physical evidence in criminal investigations. Error rates for eyewitness identifications in murder-rape cases, in contrast, may approach ten percent.

But the accuracy of DNA testing should be measured in errors per thousand cases, not errors per trillion cases as frequently asserted by FBI DNA experts in criminal trials. The discrepency probably arises from a factually incorrect assumption that there is complete statistical independence of genetic markers in real life human beings. Even slight deviations from complete independence greatly increases the likelihood of coincidental matches.

This knowledge counsels against convictions made on bare DNA evidence without corroborating evidence.

This also counsels against including people with no criminal records in DNA databases which would dramatically increase the risk of false positive results. In a DNA database including everyone in the United States, it wouldn't be surprising if 50% or more of the population has someone with a matching DNA profile using the small number of markers involved; the vast majority of whom have never engaged in serious criminal activity in their lives. Many people, no doubt, would have multiple such matches. A bigger database would certainly make the process of narrowing down suspects swift and more reliable, because it would catch first time offenders who weren't in the database (often in violent crimes where there is significant physical evidence) who wouldn't otherwise be suspects who could be given DNA tests. But the more DNA profile matches there are out there, the more likely it is that someone who was in the same physical vicinity of the crime but had no involvement in it, will come up as a match and be convicted on that basis.

There is also significant fundamental risk of false negatives. One source of this risk is people who are chimeras. Simply put, some people, usually due to a failed twin pregnancy or shared tissue during a pregnancy, have different DNA in different parts of their body. If the evidence and the DNA test sample come from different parts of the same person's body, the result could be a false negative. According to New Scientist magazine, there are about 30-40 documented human cases in the literature, several of which of surfaced in court cases. The actual frequency of this condition is impossible to know, however, because natural chimeras are almost never detected. There are very few symptoms of this condition that can be ascertained without DNA testing and those symptoms that are visible in some cases such as male/female or hermaphrodite characteristics or skin discolouring can have multiple causes some of which are more likely explanations for each condition.

Another fundamental cause (as opposed to an experimental one) for a false negative could be a germ line virus. Simply put, some viruses change your DNA when you are infected with them in a way that is permanent and passed on to future generations. About 8% of the human genome has a viral source. At least one such virus, for example, is transmitted by insect bites and found with some frequency in parts of South America. If the virus changed one of the markers used in the DNA test, a pre-infection DNA test from an individual could cease to match their current DNA test results. Since germ line viruses are not easily distinguishable from other viruses and don't carry any obvious long term symptoms, there is no way to estimate the frequency with which people are infected with them. But, even a tiny percentage rate of germ line virus infection could dramatically reduce the a theoretical false negative rate for DNA testing.

There also isn't good data on experimental error in DNA testing. Given the high pressure, high volume, low budget nature of many government run DNA testing labs, and an inherent incentive to bias in favor of the proescution that employs them the vast majority of the time, there are surely mistakes sometimes. The better studied area of medical mistakes shows that even doctors with the best intentions make serious mistakes with surprising frequency, and it is safe to assume the experimental errors in DNA testing are in the cases per thousand and not the cases per billion or trillion order of magnitude as well.

Moreover, the evidence from the area of medical mistakes is the some doctors are dramatically more likely to make serious medical mistakes than other doctors. So, even if a DNA testing lab, overall, is quite accurate, cases that happen to be assigned to the least careful person in the lab may have a far higher chance of being inaccurate than the gross error rate for the lab would suggest. We've seen waves of cases challenged due to the shoddy work of one bad expert before, and such cases are likely to come up from time to time in the future unless adequate safeguards are put into the process to insulate against individual techician incompetence.

2 comments:

Cliff dna testing labs said...

Yes we definately should place more regulation in the lab, they are under pressure to perform but they have to be accurate

Anonymous said...

I was thinking about percentages the other night and how the numbers go up with the higher the main number goes up. From their I started thinking of DNA and the possible flaws. So if you were in a courtroom and someone said that the DNA says that it is 99.9 percent positive that you are a match, given the population of the US estimated at 304 million in 2008 that would leave ane error margin of 304 thousand people that could have also been in your position. I found this scary for both sides innocent and guilty. Because 304 thousand divided by 50 means that on average there could be 604 people in any given state to meet the same criteria. Then take into account the word average what is to say that the 304 thousand number is not located in the same state? I don't know about you but it made me think more about 99.9 percent sure!!!!