On some surprising statistical properties of a DNA fingerprinting technique called AFLP

AFLP is a widely used DNA fingerprinting technique, resulting in band absence - presence profiles, like a bar code. Bands represent DNA fragments, sampled from the genome of an individual plant or other organism. The DNA fragments travel through a lane of an electrophoretic gel or microcapillary system, and are separated by length, with shorter fragments traveling further. Multiple individuals are simultaneously fingerprinted on a gel. One of the applications of AFLP is the estimation of genetic similarity between individuals, e.g. in diversity and phylogenetic studies. In that case, profiles of two individuals are compared, and the fraction of shared (comigrating) bands is calculated, e.g. using the Dice similarity coefficient. Two comigrating bands may share the same fragment, but band sharing could also be due to chance, if two equally sized, but different fragments are amplified. This is called homoplasy. Homoplasy biases similarity coefficients. Homoplasy could also occur within a lane, if two different fragments of equal length are amplified, resulting in a single band. We call this collision. The main objective of this thesis is the study of collision and homoplasy in AFLP. The length distribution of AFLP fragments plays an important role. This distribution is highly skewed with more abundant short fragments. By simulation the expected similarity for unrelated genotypes is calculated. As much as 40% of the bands may be shared by chance in case of profiles with 120 bands. The collision problem is analogous to the birthday problem, which has a surprising solution. The collision problem is even more extreme, making it even more surprising. Profiles with only 19 bands contain collision(s) with probability 1/2. These findings have consequences for practice. In some cases it is better to prevent the occurrence of collisions by decreasing the number of bands, in other cases a correction for homoplasy and collision is preferred. Modified similarity coefficients are proposed, that estimate the fraction of homologous fragments, correcting for homoplasy and collision. Partially related to homoplasy and collision, we study the codominant scoring of AFLP in association panels. Examples of AFLP in lettuce and tomato serve as illustrations.