Progress 09/01/23 to 08/31/24
Outputs
Target Audience:1) Scientific audiences in statistics, applied mathematics, epidemiology, network science, infectious disease modelling, and agricultural economy. 2) Practitioners in industry, especially hop and cucurbit producers and allied industry technical support personnel. 3) Government agencies charged with managing SOD, WNV, and H151. 4) General citizenry with interests in disease spread and mitigation. Changes/Problems:The major challenge with the SOBOL process is that it is highly computationally expensive. We need to run 320,000 simulations for our uncertainty analysis with SOD. The method scales by a factor of two for each unique uncertainty source partitioned. What opportunities for training and professional development has the project provided?A graduate student at Oregon State University received training on statistical methods supporting Goal 1. A technician at Oregon State University received training in conducting field studies on the epidemiology of wheat stripe rust in support of Goal 1. A postdoctoral research associate at North Carolina State University received training on experimental design, molecular methods, and statistical methods supporting Goals 1 and 2. An Adjunct Lecturer and Junior Researcher at City University of New York received training on optimal control policy modeling by simulation. A PhD Student at Oregon State University Oregon State University received training on statistical methods supporting Goals 1 and 4. A PhD student at North Carolina State University is being trained on implementing the SOBOL methodology. A second student is working on citrus black spot to verify that our methodology works across systems. Both students also received Fellowships with the Pacific Northwest National Laboratory to expand the uncertainty analysis done for SOD using our process-based model to a machine learning model for the spread of Avian Influenza. Three PhD students at Kansas State University were partially supported by this project and are becoming knowledgeable on stochastic models of infectious disease, model identification, and WNV modeling literature. A post-doc at University of Warwick received training on modeling of disease spread and spillover effects in Highly Pathogenic Avian Influenza (H5N1). How have the results been disseminated to communities of interest?1) An outreach presentation was made to cucurbit industry stakeholders. 2) We are working with the Oregon Department of Forestry (ODF) and the USFS to use our results as part of the decision-making process for managing sudden oak death (SOD). Together, we have identified the next steps for using the results of this analysis in an iterative forecasting workflow for SOD. 3) We used data collected from commercial hop yards to develop a linked economic-epidemiological model that estimates the incidence of powdery mildew on leaves and cones, crop damage from the disease, the number of fungicide applications made, their costs, and resulting profit. The model considers outcomes due to disease develop from both inoculum produced within a given field and inoculum that may be dispersed from the network of all other yards in the landscape. 4) We conducted simulation studies of hop powdery mildew to understand how factors related to initial pathogen population, pathogen diversity, the location of primary infection, and management influence profitability and economically optimal outcomes. The model indicates that profitability considered in aggregate at the landscape level is sensitive to the frequency of overwintering of the pathogen, frequency of virulence in the pathogen population, market conditions as they affect crop devaluation, and the disease management strategy growers adopt (i.e., the number of fungicide applications made). Analyses are largely completed, and we began drafting a manuscript describing the model and simulations results. This work provides a framework of how to synthesize fundamental knowledge of pathogen dispersal with situated economic conditions to determine management strategies that are optimized at a regional scale for a population of producers. Our result point to intervention strategies targeting pathogen overwintering and where host resistance might be most impactful when it is deployed to reduce the severity and economic impacts of epidemic. 5) Research findings were presented at two international meetings of scientists and a seminar at Cornell University. 6) Two scientific journal articles were published, one has been accepted, two are in a second round of revisionon, and one has been accepted; others are in preparation. What do you plan to do during the next reporting period to accomplish the goals?Goal 1: a. Optimize field experiment design, conduct sampling, and complete data collection for cucurbit downy mildew. b. Field experiments have been planted to repeat wheat stripe experiments for effects of number of outbreak foci. Plots will be inoculated and disease spread data collected. c. Use simulated data and experimental data of both cucurbit downy mildew and wheat stripe rust to test and validate the source detection model. Goal 2: a. Optimize field experiment design, conduct sampling, and complete field data collection for cucurbit downy mildew. b. Complete data collection to confirm pathogen lineage present on each cucurbit host. c. Build a framework to model the spillover effects for cucurbit downy mildew. d. Continue developing models for spillover using network models. We will be looking for data of any disease system presenting spillover and for West Nile virus specifically. Goal 3: a. We will finalize our analysis of the uncertainty estimation for the SOD system and publish the results, release the updated version of PoPS with this implementation, and use our results for targeted data collection using remote sensing and machine learning to improve our coverage of initial conditions (i.e., locations with positive SOD identification). b. Work on assessing disease spread models uncertainty using network models, with an emphasis on WNV. Goal 4: a. Continue working on concepts and models to unify framework of biological processes across LDD systems using multiple approaches.
Impacts
What was accomplished under these goals?
1. Determine if locations of multiple sources of epidemic outbreak can be imputed from population dynamic models. a. Field experiments were designed and implemented to create experimental epidemics initiated with one or more epicenters using cucurbit downy mildew as a model system of an aerially dispersed pathogen. b. Field experiments were successfully conducted with wheat stripe rust to compare epidemics initiated with one, two or four outbreak foci. c. Spatio-temporal models incorporating anisotropic spread induced by wind have largely been coded and implemented. We are now working on a Bayesian implementation of the model to identify the source locations given a known disease spreading process and observations of disease. These models will be used to determine ability to impute locations of initial outbreak sites for both the cucurbit down mildew and wheat stripe rust epidemics. 2. Evaluate effects of ecological spillover on the spread of LDD epidemics and the observed relationships between disease spread and taxonomic diversity of hosts. a. Field experiments were designed and implemented to create epidemics of cucurbit downy mildew initiated by mixed pathogen populations in host populations that vary in the diversity of a primary versus secondary host to model factors driving spillover effects. Disease data have been collected, cleaned, and quality assured. We collected lesions from host plants in each replicate plot, extracted DNA, and assayed each sample by quantitative PCR to determine the presence and relative quantity of each of two clades of the cucurbit downy mildew pathogen on each host. Data analyses are planned to model spillover as multilayered network with edges inferred by host density, wind-driven dispersal, and pathogen source strength. b. The Warwick team has developed a farm-based kernel model for the UK poultry system, utilizing data from the recent outbreaks of Highly Pathogenic Avian Influenza H5N1 in the country. This model has been fitted to outbreaks from October 2021 to October 2023 and has been used to assess the relative impacts of spillover from the wild bird population and within industry transmission. The model has been developed in a flexible manner such that it can also be utilized to simulate cattle diseases such as foot and mouth disease by incorporating the demographic data on cattle farms. c. We have added specifics into the PoPS simulation model to separate host locations and host parameters to understand what aspect of the host is driving transmission. Our main question is whether the location or the underlying biology (e.g., susceptibility and competency of the host in relation to the pathogen) leads to areas with higher spillover. We will test these questions through computational experiments as we are unable to do inoculation experiments for SOD. e. We examined two distinct scenarios involving the dynamics of Susceptible-Infected-Recovered (SIR) in interconnected networks. In the first part, we showed how the epidemic threshold of a contact network changes as a result of being coupled with another network for a fixed infection strength. In the second part, we investigate the spillover phenomenon, where the disease in a novel host population network comes from a reservoir network. We have observed a clear phase transition when the number of links exceeds a certain threshold, keeping all other parameters constant. f. Concerning spillover in West Nile virus, we have introduced a novel probabilistic approach for assessing the risk of West Nile virus spillover to the human population. In contrast to other methods, which do not allow for long-term risk assessment, this method first enables us to predict the risk over a long period (for example, one year) and second, it improves the accuracy of the prediction as time progresses. 3. Evaluate the effects and sources of uncertainty needed to obtain robust predictions of pathogen transmission. a. We have implemented a SOBOL methodology for partitioning uncertainty both spatially and temporally across four key model areas: drivers (host and weather), processes (stochastic process in the model), parameters (parameters are estimated using Approximate Bayesian Computation), and initial conditions (initial detections). We applied our methodology on sudden oak death (SOD) using the PoPS (Pest or Pathogen Spread) modeling framework in Curry County, OR for a five-year forecast. Our results show that uncertainty in initial conditions (i.e., the incomplete spatial coverage of SOD detection) is the largest source of uncertainty in the wavefront of the SOD outbreak. We are working on releasing this methodology as a new tool in the PoPS modeling package and a stand-alone package that can be used with other models and other pathosystems. b. We have started a detailed study to understand the true meaning of uncertainty, its main sources, and the best ways to address it. Uncertainty, as a natural part of complex systems, significantly influences behavior and outcomes. Rather than seeing it only as a challenge, it serves as an important source of insight that can guide research and highlight knowledge gaps. This idea includes various sources of uncertainty, such as inherent randomness in processes and limits in knowledge or modeling assumptions. By systematically identifying these sources, we can create a complete framework for uncertainty quantification. c. We detailed a real-time prediction system for West Nile Virus (WNV) that incorporates an adapted compartment model to account for the transmission dynamics among birds, mosquitoes, and humans, including asymptomatic cases and the influence of weather factors. To reduce data uncertainties, assimilation methods are used to integrate new surveillance data into the epidemiological model to update predictions. Techniques such as the ensemble Kalman filter (EnKF) and the particle filter (PF) are commonly used for this effort. The EnKF, by updating predictions with each new piece of information, significantly refines the accuracy of short-term forecasts and contributes to a more comprehensive understanding of disease transmission dynamics. Using the EnKF, we generated weekly WNV case forecasts for Colorado in 2023, providing valuable insights for public health planning. Comparative analyses underscore the enhanced forecast accuracy achieved by integrating weather variables into our models. 4. Develop a unifying framework of biological processes across LDD diseases incorporating diverse hosts, pathogens, and environments evolving with time. a. Predicting the spread of processes across complex multi-layered networks has long challenged researchers due to the intricate interplay between network structure and propagation dynamics. Our study introduces a novel framework that efficiently predicts Markov processes over large-scale networks, overcoming the state explosion problem common in Markov processes while significantly reducing time and space complexity. This approach enables the simulation of spreading processes across extensive real-world multilayer networks, accounting for diverse influencers on each layer. This framework opens new avenues for understanding and predicting complex spreading phenomena in many fields of computational epidemiology promising to revolutionize our ability to model and manage large-scale network interactions.
Publications
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Das, S., M. Samaei, and C. Scoglio. 2024. SIR epidemics in interconnected networks: threshold curve and phase transition. Applied Network Science. 9:50.
- Type:
Peer Reviewed Journal Articles
Status:
Published
Year Published:
2024
Citation:
Seibel, R.L., Meadows, A.J., Mundt C., and Tildesley M. 2024. Modeling target-density-based cull strategies to contain foot-and-mouth disease outbreaks. PeerJ 12:e16998.
|
Progress 09/01/22 to 08/31/23
Outputs
Target Audience:International communities of plant health scientists, epidemiologists, and disease ecologists. Industry stakeholders, agricultural producers and their advisers for livestock, cucurbits, wheat, and hops. Government agencies charged with managing sudden oak death. Government agencies charged with managing avian influenza and foot-and-mouth disease. Undergraduate and graduate students in epidemiology, ecology, and quantitative biology. Changes/Problems: a. In the Fall season, there was natural infection of cucurbit downy mildew in the field. Therefore, artificial inoculation was not conducted. Given the pre-existing disease, this reduced the number of factors that could be varied experimentally, such as location of disease foci. In the future, we will time experiments to better allow for artificial inoculation. b. The major challenge with the Sobol process for evaluating sources of stochasticity is that it is highly computationally expensive. We need to run 320,000 simulations for our uncertainty analysis with SOD. The method scales by a factor of two for each unique uncertainty source partitioned. This issue is being addressed by running simulations continuously and by searching for availability of machines with improved computational speed. c. For wheat stripe rust, spring and late winter were unusually cold but was followed by an unusually warm and dry late spring/early summer. As a result, we were able to obtain data from only two generations of disease spread and lost data from a few plots due to crop maturity. These are highly unusual conditions for the Willamette Valley and unlikely to reoccur. In addition, personnel came on the project late into the first year of the project and we may thus have resources to run field experiments for another year with a no-cost extension if need be. What opportunities for training and professional development has the project provided? a. A postdoctoral research scholar is being trained in epidemiology of the cucurbit downy mildew system at North Carolina State University b. A graduate student at Oregon state university is working with data sets regarding the spread of hop powdery mildew. c. Two new PhD students at Kansas State University have been supported by this project and are becoming knowledgeable on stochastic models of infectious disease, model identification, and West Nile Virus modeling literature. d. A PhD student at North Carolina State University is being trained in evaluating sources of variation in the sudden oak death system and and implementing the Sobol methodology. e. An undergraduate at North Carolina State University is working on citrus black spot. f. Joshua Pedro, Lecturer at City University of New York, worked on broader impacts part of the project. g. A faculty research assistant has developed skills in quantifying the spread of wheat stripe rust at Oregon State University. h. A postdoc at University of Warwick gained experience if developing predictive models for predictive models of avian influenza that also will be used for foot-and-mouth disease of livestock. How have the results been disseminated to communities of interest? a. Commonalities in the long-distance spread of plant and animal diseases were presented in the graduate class "Plant Disease Dynamics" at Oregon State University. b. A presentation was given on spread of cucurbit downy mildew at the Predicting the Next Plant Disease Pandemic Symposium, sponsored by the NSF Predictive Intelligence for Pandemic Prevention Phase II Program c. A manuscript on disease spillover has been submitted to the Springer journal "Applied Network Science," and another is in progress. d. We have interacted with government agencies charged with managing sudden oak death.? What do you plan to do during the next reporting period to accomplish the goals?1. Locations of multiple sources of epidemic outbreak can be imputed from population dynamic models. Optimize cucurbit downy mildew field experimental design and finish the experiment before August. Repeat the wheat stripe rust experiment in the field. Use simulated data and collected real data to test the existing model and find a way to upgrade the disease source detection model. 2. Ecological spillover effects influence the spread of LDD epidemics and the observed relationships between disease spread and taxonomic diversity of hosts. a. Complete data collection to confirm pathogen lineage present on each host in the cucurbit field study. b. Analyze spillover effects between two cucurbit hosts using the severity data and lineage information. c. Build a general framework and develop a general model spillover effects. d. Collect pertinent data and begin work on to evaluate spillover effects for West Nile Virus. e. Begin analyzing spillover effects in simulation studies of foot-and-mouth disease of livestock. f. Begin analyzing spillover effects in simulation studies of sudden oak death.. 3. Robust predictions of pathogen transmission require understanding the effects and sources of uncertainty. a. Begin work on assessing uncertainty in disease spread models for West Nile Virus. b. Finalize uncertainty estimations for sudden oak death, release an updated version of the PoPS model with this implementation, and publish this work based on the sudden oak death system. 4. A unifying framework of biological processes emerges across LDD diseases incorporating diverse hosts, pathogens, and environments evolving with time. a. Begin a general comparison of results across disease systems and identify questions to be addressed using modeling and other approaches. Broader Impact Goals: 1) Identify improved management strategies for important animal and plant diseases. a. Finish code for framework of generic predictive model for use with real-time data, parametrize the model for avian influenza, and begin application to foot-and-mouth disease. This model will then be used with real time data to carry out forward simulations of the impact of control. 2) Investigate optimization of disease mitigations in terms of policies, societal impacts, and epidemic spread. a. Complete simulations to link epidemic and management conditions to economic outcomes for the hop powdery mildew pathosystem. b. Share results of the above modeling with communities of interest at scientific conferences and industry outreach events.
Impacts
What was accomplished under these goals?
1) Locations of multiple sources of epidemic outbreak can be imputed from population dynamic models. a. For cucurbit downy mildew, randomized and replicated complete block experiment with five treatments (a factorial combination of initial focus size and number and a non-inoculated control) was conducted during May to July 2023. The plots were artificially inoculated and disease assessments were conducted twice per week from June 30 to July 17. Initial data visualizations indicate strong disease gradients due to pathogen dispersal over time and space relative to the placement and number of the initial disease foci. Data analysis is ongoing to build a Bayesian modeling framework to estimate the posterior probability of source locations given a disease spread model and observed patterns of disease spread over time. b. A wheat stripe rust experiment was conducted to produce a data set for evaluating whether multiple outbreak sources can be imputed from disease spread data. All plots consisted of the same susceptible cultivar in 24.4 x 24.4 m plots with 24.4 m between plots and a 10.7 m border around the edges of the fields. Interplot area and field edges were sown to a highly resistant cultivar. Plots were artificially inoculated to establish outbreaks beginning from differing numbers of outbreak foci. There were three treatments: one outbreak focus, two outbreak foci, and four outbreak foci. Each treatment was replicated four times and the spatial location of the inoculated foci was randomly chosen independently for each replication of each treatment. All inoculation sites were 0.46 x 0.46 m. Disease severity data were obtained in a grid pattern over the plots and at different points in time. Data are still being evaluated, though it appears that different outbreak sites are usually discernable. Though not one of our original goals, results also suggest a very large effect of number of outbreak sites on overall disease levels in the plots. c. We began analyses utilizing extant data on hop powdery mildew occurrence and spread at the mesoscale. We implemented an existing disease spread model for powdery mildew with this data set and intend to modify this model to similarly estimate posterior probabilities of (known) initial disease foci in the landscape. 2) Ecological spillover effects influence the spread of LDD epidemics and the observed relationships between disease spread and taxonomic diversity of hosts. a. We successfully conducted field experiments with cucurbit downy mildew to enable modeling of pathogen spillover as related to host density. We utilized butternut squash (S) and cucumber (C) to represent the primary and secondary hosts of the pathogen, respectively. A randomized and replicated incomplete block experiment that was designed to vary the proportion of each host and size of the initial disease focus (50% S 50% C with a small focus, 70% S 25% C with a small focus, 50% S 50% C with a big focus, 70% S 25% C with a large focus, and non-inoculated S or C controls) with three replications was conducted from August to October 2023. Natural infections occurred prior to artificial inoculation and therefore the initial size of the disease focus could not be varied experimentally. From September 25 to October 6, a total of four disease assessments were conducted twice a week. At the same time, leaf discs were collected from diseased plants for confirmation of which of two host-adapted pathogen strains (lineages) were present on each host. In both the treatments 50% S 50% C and 75% S 25% C treatments, disease severity on squash was higher than cucumber during all disease assessment times. Moreover, the severity on cucumber was lower in the treatment of 75% S 25% C than the treatment of 50% S 50% C. Disease severity on squash was not higher than cucumber in the control plots in the first two assessments, while severity on squash was slightly higher than that of cucumber in the last two disease assessments. b. We have begun studying theoretical models of spillover using network-based models. We have examined two distinct scenarios involving the dynamics of Susceptible-Infected-Recovered (SIR) in interconnected networks. In the first part, we show how the epidemic threshold of a contact network changes because of being coupled with another network for a fixed infection strength. In the second part, we investigated the spillover phenomenon, where the disease in a novel host population network comes from a reservoir network. c. To study spillover in West Nile Virus, we are searching for a model providing a satisfactory tradeoff between complexity, by the number of species represented in the model, and obtainable accuracy. For West Nile Virus, the mechanism of spread relies on the interactions between birds and mosquitoes, with occasional transmission to humans as spillover cases. We have established our models using three types of mosquitoes, two species of birds, and a human population. One of the bird species serves as a local transmission agent, while the other acts as a long-distance transmission vector, being migratory birds.The primary goal of this work is to investigate whether the abundance of mosquito species significantly impacts the predictive accuracy of the models for assessing disease risk and predicting the number of new human cases representing spillover. d. We have added specifics into the PoPS model of sudden oak death to separate host locations and host parameters to understand what aspect of the host is driving transmission: is it the location or the underlying biology (e.g., susceptibility and competency of the host in relation to the pathogen). 3. Robust predictions of pathogen transmission require understanding the effects and sources of uncertainty.?a. We are performing a literature review on the topics of spillover in West Nile Virus and uncertainty quantification in epidemic models. b. We have prototyped a Sobol methodology, a variance-based sensitivity analysis, for partitioning uncertainty both spatially and temporally across five key model areas: drivers (host and weather), processes (stochastic process in the model), parameters (parameters are estimated using Approximate Bayesian Computation), and initial conditions (initial detections). We are testing this methodology on sudden oak death using the PoPS modeling framework. Once the methodology has been tested it will be released in an updated version of the PoPS model R package. 4. A unifying framework of biological processes emerges across LDD diseases incorporating diverse hosts, pathogens, and environments evolving with time. We expect to make progress on this goal in the later stages of the project as it will first require completion of other goals. Broader Impacts Goals: 1. Identify improved management strategies for important animal and plant diseases. a. A new version of a livestock disease model that incorporates a Bayesian Markov Chain Monte Carlo approach is being developed. Once finished, it will enable us to have a code that is able to take real time data, parameterize our model to those real time data and then carry out forward simulations of the impact of control. The disease framework being developed is generic - it is currently being developed for highly pathogenic avian influenza but the same framework will also be able to be used for foot-and-mouth disease. 2. Investigate optimization of disease mitigations in terms of policies, societal impacts, and epidemic spread. We made substantial progress on developing a linked epidemiological and economic model that estimates disease development and spread at the landscape level, taking into account fungicide effects and expected crop damage. We coupled this epidemiological model to an economic model of expected revenue and costs. The coupled model was parameterized using data collected from a census sample of commercial hop yards in Oregon during 2014 to 2017.
Publications
- Type:
Journal Articles
Status:
Submitted
Year Published:
2024
Citation:
Das, S., and Scoglio, C. 202_. SIR epidemics in interconnected networks: threshold curve and phase transition. Applied Network Science: Submitted.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Gent, D. H, Pedro, J. F., Marsh, T. L., Chatterjee, S., and Bhattacharyya, S. 2023. Coupling an epidemiological and economic model to optimize management of hop powdery mildew at the landscape level. Proceedings of the Scientific and Technical Commission of the International Hop Growers Convention, Ljubljana, Slovenia.
|