Sizing bands on autoradiograms: a study of precision for scoring DNA fingerprints.

We replicated DNA fingerprints of snapping turtles (Chelydra serpentina) and hypervariable restriction fragments of red-winged blackbirds (Agelaius phoeniceus) to estimate the between-blot and between-lane components of variance in molecular weights of restriction fragments. Molecular weight standards were included in every lane, and bands were sized using a sonic digitizer. In both studies, a strong positive correlation was found between band size and coefficient of variation (CV; mean = 0.7%). In the DNA fingerprint study, 26% of the variance in estimates of band size was due to differences between blots, 10% due to differences between lanes on the same blot, and 64% due to error in the digitizing process. In the restriction fragment length polymorphism (RFLP) study, 16% of the variance was due to difference between lanes, and 84% to digitizing. Statistical models were developed to measure the effect of sizing error on identifying identical fragments in different lanes or on different blots, in categorizing distinct alleles, and in determining the size of bins in operational allele definitions. We suggest that the distance between bands be at least 2.8 standard deviations (SD) before they are declared different at alpha = 0.05, and 3.7 SD for alpha = 0.01. A variation in CVs strongly indicates that empirical relationships between SD and band size must be used to decide if two bands represent the same allele. Alleles must be at least 3.9 SD apart before the chance of assigning new observations in error falls below 0.05. We suggest that a minimum bin width of 16 SD is necessary before the chances of assigning a band to the wrong bin falls below 0.05.