Extract DNA from the forensic sample, the victim, and the suspect
DNA profiling is typically employed in two scenarios, in violent crimes such a murder or rape and in paternity (or grand-paternity) cases. The amount of biological samples available, whether it be blood, semen or hair, may often be old and in poor condition. Fortunately DNA is quite stable and resistant to degradation. More seriously, the amount of sample available may be very limited. However, with the latest procedures available sufficient DNA may be obtained from just a single hair follicle.

Courtesy, Lab Corp. of America

Commercial enterprises emerged to fill a public need to determine paternity of children, as seen in this humorous advertisement from Genetic Design, Inc.

In paternity cases, a blood sample (or even a cheek swab) will be collected from the child whose paternity (or grand-paternity) is in question and the child's immediate relatives, so that the questions concerning the condition or quantity of the sample are not in dispute.

In crime cases, it may frequently be the situation that no suspect has been identified. In such circumstances, an increasing number of states can resort to a DNA database of individuals who have previously been convicted of a violent crime. By 1998, around 200 "matches" had been made using such databases.

Regents of the University of Michigan

Autoradiograph from an actual rape case showing the DNA profiles for one VNTR locus. The lanes marked "M" show a "ladder" of DNA fragments of known sizes. These are loaded onto the gel to provide an internal ruler--allowing the sizes of the VNTR alleles to be estimated more accurately. As can be seen there is a match of the DNA profile of defendant 1 and the forensic sample.

One of the first successful prosecutions involving a database occurred in England in the early 1990s. A woman walking home in the early hours of the morning in an outlying London suburb was approached by a man who attacked her and raped her from behind. She was only able to catch a glimpse of his face for a few seconds in the darkness. No suspect was identified until three years later, when the DNA profile of an individual, apprehended for an unrelated offense, was found to match that of the semen sample taken from the victim. Even though the victim was unable to single out the defendant in an identity parade, he was convicted. A vigorous defense was mounted, but after a retrial and two appeals, the original verdict stood.

Determine if there is a match
We determine that there is a match if the sizes of the alleles (one of two or more alternate forms of a gene) at all the loci tested are the same in the defendant and the forensic sample. But what do we mean by "the same"? All biological experiments are subject to "experimental error"--variation that is random. Certainly sizing VNTR or STR alleles by electrophoresis is no exception. Thus, if we run the same DNA sample time after time, the migration distances of the alleles will be slightly different each time. So when we compare DNA from the forensic sample and the defendant, we will never see exactly the same positions on the gel, even if in reality the DNA from the forensic sample originated from the defendant.

We must therefore determine how much experimental error we would expect by running the same sample repeatedly, and then we use a statistical test to determine if the alleles are the same or different. The notion of a statistical test to determine similarity or difference may generate concern, as statistical tests can only say that samples are the same or different with a certain probability. That is, they are not absolute determinants of similarity or difference, and consequently there is error associated with any test. We can imagine two kinds of errors in DNA testing, which are related.

Most would agree that false positive errors are unacceptable, while false negative errors, while unfortunate, are acceptable providing they occur seldom. In fact the statistical test for similarity is designed to be quite stringent to minimize the risk of false positives. However an exceptionally stringent test will increase the chance that false negative errors will occur--that is we will mistakenly declare no match.

	Discussion
	At least nine former death row inmates in the US have been exonerated because of DNA testing. What margin of error (for false positives and false negatives) should be acceptable in the eyes of the law for cases of violent crime?

Imagine that we just accept as matches, alleles that migrate exactly the same distance in the electrophoresis process. The possibility that we will mistakenly declare a match (false positive), when none exists, will be vanishingly small. However, we will mistakenly declare that there is no match (false negatives) frequently as the same alleles will show small variations in migration distance due to experimental error. Now imagine that we accept as matches all alleles that have migration distances within two inches of each other--an absurdly large tolerance range as the electrophoretic gel may be as small as 8 inches in length. Then we can imagine that we will mistakenly declare a match (false positives) often. On the other hand the numbers of false negatives will be vanishingly small. Thus we can see the two types of errors are related. We cannot minimize one without adversely affecting the other.

The use of many loci also minimizes the possibility of false positive error. To declare a match with 10 loci, all 20 alleles (two/locus) must match. If just one allele is different in the forensic sample and the defendant, then there is no match.