Development of statistical models for prediction of the neurotoxin domoic acid levels in the pennate diatom Pseudo-nitzschia multiseries utilizing data from cultures and natural blooms

From National Research Council Canada

AuthorSearch for: ; Search for: ; Search for: 1; Search for: ; Search for:
Name affiliation
  1. National Research Council of Canada. NRC Institute for Information Technology
FormatText, Book chapter
Book titleAlgal Cultures, Analogues of Blooms and Applications
ISBN9781578085620
Pages891–916
Abstract
Domoic acid (DA), a neurotoxin implicated in amnesic shellfish poisoning episodes, is produced by the pennate diatom Pseudo-nitzschia multiseries and its presence is associated with natural blooms of this marine microorganism. Laboratory studies indicate that production of the toxin by the diatom is related to physiological stress caused by limitation of nutrients, such as phosphate and silicate. This study attempts to develop predictive models of the amount of DA present, using data from laboratory physiological studies of the diatom, and from a combination of the laboratory data with field data from one Atlantic and two Pacific Ocean sites between 1988 and 1998. Such models can provide early warning of a potential poisoning episode without the sophisticated chemical determinations of domoic acid and avoid delays due to lack of analytical facilities and expertise. Predictor variables include nutrient ratios and cell abundance of Pseudo-nitzschia multiseries, determinations of which are both simple and inexpensive. Both linear and logistic regression methods are employed. Natural logarithm and square root transformations of predictors and response variables are needed to linearize the modeled relationship. In total, four models are proposed. Three of these models have r 2 ranging between 0.60 to 0.80 and are based on between four and seven predictors. Split sample reliability testing indicated that shrinkage on cross-validation for one of these models was below 35%. The fourth, a logistic regression model developed using the split-sample approach, has 2 predictors. In the training data, the prediction sensitivity and specificity of this model were approximately 76%. In the Validation data, the specificity increased to 87%, while the sensitivity declined to 67%.
Publication date
PublisherScience Publishers Inc.
Science Publishers Inc.
LanguageEnglish
Peer reviewedYes
NPARC number23002084
Export citationExport as RIS
Report a correctionReport a correction
Record identifier198a4ac4-477a-41b0-9d9d-b813890d1bff
Record created2017-08-10
Record modified2020-06-16
Report a problem on this page
Date modified: