excludes a suspect, assessing the significance of a result requires statistical analysis of population frequencies.
Despite the challenges of forensic DNA typing, we believe that it is possible to develop reliable forensic DNA typing systems, provided that adequate scientific care is taken to define and characterize the methods. We outline below the principal issues that must be addressed for each DNA typing procedure.
An essential element of any clinical or forensic DNA typing method is a detailed written laboratory protocol. Such a protocol should not only specify steps and reagents, but also provide precise instructions for interpreting results, which is crucial for evaluating the reliability of a method. Moreover, the complete protocol should be made freely available so that it can be subjected to scientific scrutiny.
There must be an objective and quantitative procedure for identifying the pattern of a sample. Although the popular press sometimes likens DNA patterns to bar codes, laboratory results from most methods of DNA testing are not discrete data, but rather continuous data. Typically, such results consist of an image—such as an autoradiogram, a photograph, spots on a strip, or the fluorometric tracings of a DNA sequence—and the image must be quantitatively analyzed to determine the genotype or genotypes represented in the sample. Quantitation is especially important in forensic applications, because of the ever-present possibility of mixed samples.
Patterns must be identified separately and independently in suspect and evidence samples. It is not permissible to decide which features of an evidence sample to count and which to discount on the basis of a comparison with a suspect sample, because this can bias one's interpretation.
When individual patterns of DNA in evidence sample and suspect sample have been identified, it is time to make comparisons to determine whether they match. Whether this step is easy or difficult depends on the resolving power of the system to distinguish alleles. Some DNA typing methods involve small collections of alleles that can be perfectly distinguished from one another—e.g., a two-allele RFLP system based on a polymorphism at a single locus. Other methods involve large collections of similar alleles that are imperfectly distinguished from one another—e.g., the hypervariable VNTR