Predictive Modeling of Avian Influenza in Wild Birds

Thumbnail
Name:
Herrick Dissertation vH FINAL.pdf
Size:
3.435Mb
Format:
PDF
Download
Author
Herrick, Keiko
Chair
Huettmann, Falk
Committee
Taylor, Barbara
Runstadler, Jonathan
Ickert-Bond, Stephanie
Bortz, Eric
Keyword
avian influenza
data mining
randomforest
wild birds
Metadata
Show full item record
URI
http://hdl.handle.net/11122/1837
Abstract
Over the past 20 years, highly pathogenic avian influenza (HPAI), specifically Eurasian H5N1 subtypes, caused economic losses to the poultry industry and sparked fears of a human influenza pandemic. Avian influenza virus (AIV) is widespread in wild bird populations in the low-pathogenicity form (LPAI), and wild birds are thought to be the reservoir for AIV. To date, however, nearly all predictive models of AIV focus on domestic poultry and HPAI H5N1 at a small country or regional scale. Clearly, there is a need and an opportunity to explore AIV in wild birds using data-mining and machinelearning techniques. I developed predictive models using the Random Forests algorithm to describe the ecological niche of avian influenza in wild birds. In “Chapter 2 - Predictive risk modeling of avian influenza around the Pacific Rim”, I demonstrated that it was possible to separate an AIV-positivity signal from general surveillance effort. Cold winters, high temperature seasonality, and a long distance from coast were important predictors. In “Chapter 3 - A global model of avian influenza prediction in wild birds: the importance of northern regions”, northern regions remained areas of high predicted occurrence even when using a global dataset of AIV. In surveillance data, the percentage of AIV-positive samples is typically very low, which can hamper machine-learning. For “Chapter 4 - Modeling avian influenza with Random Forests: under-sampling and model selection for unbalanced prevalence in surveillance data” I wrote custom code in R statistical programming language to evaluate a balancing algorithm, a model selection algorithm, and an under-sampling method for their effects on model accuracy. Repeated random iv sub-sampling was found to be the most reliable way to improved unbalanced datasets. In these models cold regions consistently bore the highest relative predicted occurrence scores for AIV-positivity and describe a niche for LPAI that is distinct from the niche for HPAI in domestic poultry. These studies represent a novel, initial attempt at constructing models for LPAI in wild birds and demonstrated high predictive power.
Description
Thesis (Ph.D.) University of Alaska Fairbanks, 2013
Table of Contents
TABLE OF CONTENTS Page SIGNATURE PAGE ... i TITLE PAGE ... ii ABSTRACT ... iii TABLE OF CONTENTS ... v LIST OF FIGURES ... viii LIST OF TABLES ... x LIST OF ADDITIONAL MATERIALS ... x LIST OF APPENDICES ... xi DEDICATION ... xiii ACKNOWLEDGEMENTS ... xiv CHAPTER 1: General Introduction ... 1 Avian influenza virus, transmission, and pandemic potential ... 1 Modeling AIV ... 7 Specific aims ... 10 FIGURES ... 14 LITERATURE CITED .. 17 CHAPTER 2: Predictive risk modeling of avian influenza around the Pacific Rim ... 26 ABSTRACT ... 26 INTRODUCTION ... 28 MATERIALS AND METHODS ... 31 Data layers ... 31 Modeling methods ... 33 Model evaluation ... 35 RESULTS ... 35 DISCUSSION ... 37 ACKNOWLEDGEMENTS ... 40 TABLES ... 42 vi FIGURES ... 45 LITERATURE CITED .. 49 CHAPTER 3: A global model of avian influenza prediction in wild birds: the importance of northern regions ... 54 ABSTRACT ... 54 INTRODUCTION ... 55 MATERIALS AND METHODS ... 57 Wild bird data ... 57 Environmental variable layers ... 57 Defining the outbreak niche ... 59 Predictive map ... 60 RESULTS ... 61 Important predictor variables ... 61 Ecological niche model ... 63 DISCUSSION ... 63 ACKNOWLEDGEMENTS ... 68 TABLES ... 69 FIGURES ... 72 LITERATURE CITED .. 76 CHAPTER 4: Modeling avian influenza with Random Forests: under-sampling and model selection for unbalanced prevalence in surveillance data ... 80 ABSTRACT ... 80 1. INTRODUCTION . 81 2. MATERIALS AND METHODS ... 86 2.1 Predictor variables ... 86 2.2 Wild bird data ... 87 2.3 Random Forests, balancing, and model selection ... 89 2.4 Predictive map .... 92 2.5 Statistical analyses ... 92 2.6 Variable importance ... 93 vii 2.7 Cross-model comparisons ... 94 2.8 Research design .. 94 3. RESULTS ... 95 3.1. Model Performance ... 95 3.2. Cross-model comparison ... 97 3.3. Variable importance ... 98 3.4 Predictive map ... 100 4. DISCUSSION ... 101 4.1. Random sub-sampling and model selection ... 101 4.2. Database comparisons ... 102 4.3. Predictive map ... 103 4.4. Important variables ... 104 4.5 Conclusions ....... 105 ACKNOWLEDGEMENTS ... 106 TABLES ... 107 FIGURES ... 113 LITERATURE CITED ... 123 CHAPTER 5: General Discussion ... 130 Overview ... 131 The LPAI niche vs. the HPAI niche ... 135 Technical aspects and software ... 138 Future work ... 140 Surveillance and Adaptive Management principles ... 144 FIGURES ... 146 LITERATURE CITED ... 147 APPENDICES ... 150 viii LIST OF FIGURES Page INTRODUCTION FIGURES Figure 1.1. Pacific Rim study area and wild bird surveillance locations ... 14 Figure 1.2. Global study area and wild bird surveillance locations ... 15 Figure 1.3. Pacific Rim study area and wild bird surveillance locations ... 16 CHAPTER 2 FIGURES Figure 2.1. Map of predicted relative occurrence index of avian influenza virus (AIV) in wild birds around the Pacific Rim study area and surveillance locations .. 45 Figure 2.2. Notched box plots for important variables. ... 46 Figure 2.3. Histogram density plots for important variables ... 47 Figure 2.4. Partial dependence plots for important variables ... 48 CHAPTER 3 FIGURES Figure 3.1. Histogram density plots for important variables ... 72 Figure 3.2. Partial dependence plots for important variables ... 73 Figure 3.3. Map of predicted relative occurence index of avian influenza virus (AIV) in wild birds and surveillance locations ... 75 CHAPTER 4 FIGURES Figure 4.1. Research design ... 113 Figure 4.2. Receiver Operating Characteristic (ROC) curves for experimental methods ... 114 Figure 4.3. Mean area under the receiver operating characteristic curves (AUC) of the four different experimental methods that generated them ... 115 Figure 4.4. Cross-model comparison results ... 116 Figure 4.5. Density plots for the mean temperature in April ... 117 ix Figure 4.6. Density plots for important variables ... 118 Figure 4.7. Partial dependence plots for important predictor variables ... 119 Figure 4.8. Map of predicted relative occurence index of avian influenza virus (AIV) in wild birds and surveillance locations around the Pacific Rim study area ... 121 Figure 4.9. A conceptual diagram illustrating differences between traditional and collaborative surveillance methods and their interaction with laboratory and machine-learning work. ... 122 GENERAL DISCUSSION FIGURES Figure 5.1. Density plot of latitude. ... 146 x LIST OF TABLES Page CHAPTER 2 TABLES Table 2.1. Predictor variables used to construct model of avian influenza in wild birds ...42 Table 2.2. Normalized importance scores for top predictor variables ...44 CHAPTER 3 TABLES Table 3.1. The predictor variables used by the Random Forests algorithm to create a global prediction map for avian influenza virus in wild birds ...69 CHAPTER 4 TABLES Table 4.1. Selected examples of the prevalence of birds testing positive for avian influenza virus (AIV) from wild bird surveillance projects ...107 Table 4.2. Predictor variables used by the Random Forests to create a prediction map for AIV in wild birds .108 Table 4.3. Descriptive summary table for databases. ...110 Table 4.4. Summary table for experimental methods. ...111 Table 4.5. Descriptive statistics for databases and models. ...112 LIST OF ADDITIONAL MATERIALS Additional Materials ... CD xi LIST OF APPENDICES Page Appendix A. List of bird species in the Alaska Asia Avian Influenza Research 2005-2007 database... 150 Appendix B. List of bird species from the NIH Influenza Research Database (IRD). ... 157 Appendix C. List of bird species in the Alaska Asia Avian Influenza Research 2005-2020 database ... 157 Appendix D. List of bird species in the Canada’s Inter-agency Wild Bird Influenza survey (CIWBI) database . 169 Appendix E. Global Layers.xml: Metadata for bioclimatic, anthropogenic, and geographic data layers ...CD Appendix F. Georeferenced Bird Data.xml: Metadata for Pacific Rim model (Chapter 1), global model (Chapter 2), and the four datasets used in Chapter 3 ...CD Appendix G. Global Layers (folder): bioclimatic, anthropogenic, and geographic data layers used in the PhD thesis “Mapping Avian Influenza in Wild Birds” Datasets (subfolder) Chapter 1 flupacV5.shp ...CD Chapter 2 globfluV6.shp ...CD Chapter 3 A3IRB.shp ...CD Chapter 3 ALL.shp ...CD Chapter 3 CIWBI.shp ...CD Chapter 3 UNIQUE.shp ...CD GEM landcover 2000 (subfolder) glc2000_v1_1_Grid: landcover ...CD GEM-Metadata.pdf...CD GLC2000_legend_summary.doc ...CD Last of the Wild (subfolder) hfp_global_geo_grid: Human Footprint ...CD hii_global_geo_grid: Human Influence Index ...CD ltw_global_geo: Last of the Wild ...CD xii livestock (subfolder) glbpgtotcor (subfolder): estimated pig density ...CD glbpototcor (subfolder): estimated poultry density ...CD sedac human world popn (subfolder) glfedens10: human population density ...CD WorldClim (subfolder) alt_30s_esri: elevation ...CD bio_30s_esri: bioclimatic variables ...CD prec_30s_esri: monthly precipitation means ...CD tmean_30s_esri: monthly temperature means ...CD WWF GLWD (subfolder) euc_hydro_1k: distance to hydrologic feature ...CD GLWD_Data_Documentation.pdf ...CD Appendix H. Example Code (folder) random subsetting 07112012.R ...CD rocr_code_071012.R ....CD Partial_plots 71712.R ...CD
Date
2013-05
Publisher
Veterinary Research
Type
Thesis
Citation
Herrick KA, Huettmann F, Lindgren MA: A global model of avian influenza prediction in wild birds: the importance of northern regions. Vet Res, in press
Collections