Michele Masuda

Classical Discriminant Analysis, Classification of Individuals, and Source Population Composition of Mixtures

Publisher Summary Classical discriminant and classification analysis has been used since the late 1950s to assess the unknown source population proportions composing fish mixtures. Measurements of individuals in learning samples from the separate sources are used to determine rules, or discriminant functions, by which to classify measured individuals of unknown sources. Linear discriminant functions are based on the assumption that the measurement distributions in the source populations are multivariate normal with common covariance matrix, or else their distributions are unspecified, but the log odds between populations are assumed to be linear. The source labels of the mixture individuals are usually assigned with uncertainty because the measurement distributions of the sources overlap. Among mixture individuals with matching measurements, the proportions composing the mixture from the separate sources limit the certainty with which they can be classified. These proportions, which vary with the measurements, are called the posterior source probabilities of the mixture individuals. The Bayes classifier, which has the lowest misclassification rate, assigns any unlabeled mixture individual to the source for which the posterior source probability is greatest.

Use of genetic data to infer population-specific ecological and phenotypic traits from mixed aggregations

Many applications in ecological genetics involve sampling individuals from a mixture of multiple biological populations and subsequently associating those individuals with the populations from which they arose. Analytical methods that assign individuals to their putative population of origin have utility in both basic and applied research, providing information about population-specific life history and habitat use, ecotoxins, pathogen and parasite loads, and many other non-genetic ecological, or phenotypic traits. Although the question is initially directed at the origin of individuals, in most cases the ultimate desire is to investigate the distribution of some trait among populations. Current practice is to assign individuals to a population of origin and study properties of the trait among individuals within population strata as if they constituted independent samples. It seemed that approach might bias population-specific trait inference. In this study we made trait inferences directly through modeling, bypassing individual assignment. We extended a Bayesian model for population mixture analysis to incorporate parameters for the phenotypic trait and compared its performance to that of individual assignment with a minimum probability threshold for assignment. The Bayesian mixture model outperformed individual assignment under some trait inference conditions. However, by discarding individuals whose origins are most uncertain, the individual assignment method provided a less complex analytical technique whose performance may be adequate for some common trait inference problems. Our results provide specific guidance for method selection under various genetic relationships among populations with different trait distributions.

The Effects of Violating Hardy–Weinberg Equilibrium Assumptions on a Cluster-based Population Mixture Analysis of Steelhead Populations in Southeast Alaska

Abstract Clustering methods for population mixture analysis assign individuals probabilistically to populations based on their multilocus genotype data. An assumption of the methods is that loci satisfy Hardy–Weinberg equilibrium (HWE) conditions within populations. We observed that violating this assumption by including loci measured as deviating from HWE in baseline samples for the mixture analysis at times introduced extra structure into the mixture sample, leading to biased composition estimates and overestimation of the number of populations. Provided that samples from at least some contributing populations are available and that baseline samples can safely be assumed to come from single populations, then a conservative approach to mixture analysis would be to include only those characters that conform to HWE in the baseline samples, but this approach could result in a loss of resolving power. To address this problem, we outlined an ad hoc method of selecting loci for their use in mixture analysis ba...

A Five-Year, In Situ Growth Study on Shallow-Water Populations of the Gorgonian Octocoral Calcigorgia spiculifera in the Gulf of Alaska

Gorgonian octocorals are the most abundant corals in Alaska where they provide important structural habitat for managed species of demersal fish and invertebrates. Fifty-nine gorgonian species have been reported from Alaska waters but little is known about their life history characteristics to help us gauge their ability to recover from seafloor disturbance. Colonies of the holaxonian Calcigorgia spiculifera were tagged beginning in 1999 at three sites in Chatham Strait, Southeast Alaska, using scuba and their growth measured annually for up to 5 years. Colonies were video recorded, and computer image analysis tools provided calibration of video images for measuring the length of several branches. Growth data indicate that C. spiculifera grows much slower (6.0 mm yr-1) than other gorgonians in Alaska for which there are data and that intraspecific growth is highly variable. We fit a Bayesian linear mixed-effects model that showed that average colony growth was significantly reduced with warmer temperature and presence of necrosis. The model further indicated that growth may slow among larger (older) colonies. Based on these results and previous studies, we propose that gorgonian growth rates are taxonomically constrained at the Suborder level and that holaxonians grow the slowest followed by scleraxonians and calcaxonians (2–3 times as fast). Findings of this study indicate that it would take approximately 60 years for C. spiculifera to grow to its maximum size and depending on the location and size of the parental standing stock, at least one and possibly 10 additional years for recruitment to occur. Our results further indicate that colonies that are injured, perhaps chronically in areas of frequent disturbance, grow at slower rates and if the current trend of ocean warming continues then we can expect these corals to grow more slowly, and the habitats they form will require more time to recover from disturbance.