INTEGRATING DATA STREAMS FOR CAUSAL INFERENCE AND FORECASTING APPLICATION TO FOSTER PRECISION SWINE HEALTH

<b>Project Methods</b><br> Objective #1 (Research): Development of the automated Predictors of Swine Performance(PROSPER) platform:Overall system design: This objective will consist of evolving the current beta PROSPER into a fully automated platform, by three steps:• "1.A" – Creating the PROSPER platform inside Boa, a well-vetted cyber infrastructure (42), housed at the Iowa State University Computer Sciences department (http://boa.cs.iastate.edu). Boa will be connected to a server in each swine production company through application programming interfaces (APIs) and will import the data streams relevant for this project on near real-time basis. The Boa cyber infrastructure has the capability to work with different large repositories, including genomics and Covid-19. For the purpose of this project, we will utilize Boa by creating a dedicated infrastructure for the PROSPER platform, allowing secure and separate raw data storage and web-access for data analysis. Storing the data inside the Boa infrastructure will ensure security and confidentiality of raw data for the swine companies enrolled on the project, and rapid data management and analysis.• "1.B" – For the purpose of analysis, the multiple data streams available from each production will be standardized, cleaned, organized, and consolidated into a single master-table, merging the data longitudinally for each group marketed pigs, i.e., the mater-table will match and merge all retrospective information that occurred for each group of pigs marketed, creating a sort of "background check" report from data collected across the 6 months from birth-to-market. This process will be will conducted using SAS and R algorithms by the Project Coordinator, which will access the PROSPER platform inside Boa through API connection with a desktop computer, running all the algorithms developed for creating the master-table for each swine production system. Initially, each master-table will contain 3 to 5 years of retrospective data of productivity, health, management, environment, facility, feed nutrition, and diagnostic results, organized for each group of marketed pigs, serving then as the foundation for further analysis.• "1.C" – The algorithms developed to create the master-table will be adapted to import new data merging with existing data. Two reports will be extracted in the format of CSV file for each participating swine production company, for further statistical analysis purposes as described in objective 2. Firstly, a monthly report of all the closeouts recently marketed, summarizing the standardized retrospective data. Secondly, weekly automated reports will be generated containing retrospective data on the most recent weaned groups of pigs, with information about their performance in the pre-weaning phase.Objective #2 (Research): Implement regression and machine learning algorithms, taking full advantage of the digital platform built in objective #1, for identifying and quantifying the causal effect of the major drivers of swine mortality under field conditions, and forecast the impact of these drivers of performance of commercial swine populations.Experimental design: The analyses conducted in this objective will be organized into three subobjectives. First, a retrospective analysis of the consolidated data will identify and measure the variables associated of swine mortality. Then, a prospective analysis will forecast the productivity of growing pigs (i.e., 5-6 months prior to slaughter). Finally, causal inference models will be constructed to reveal factors impacting the performance of marketed groups. The analyses will be conducted for each participating production system, and a benchmark analysis will be conducted with aggregated and anonymized data for sharing the results with the US swine industry.Objective #3 (Research and extension): Benchmark the major drivers of performance under field conditions over time.Study rationale and design: Benchmarking is an effective strategy to allow producers to understand where improvements can be made in their operation. At the present time, benchmarking drivers of swine performance is difficult due to the lack of standardized databases including data on swine health, productivity, and environmental conditions in which they were exposed from breeding to farrowing to weaning to nursery to finish (whole-herd). Here, we will utilize the data collected in the previous objectives to provide benchmarking data to the swine industry. For each production system enrolled in the study, we will provide a within-production system benchmarking analysis (i.e., comparison between their farms), as well as regular on-site meetings for evaluating the data collection process for further improvement of the model.Objective #4 (Extension and outreach): Establish the PROSPER' Producer support &communications team'.The major goal of this objective is to establish an effective team to ensure full connectivity and fluidity between the research objectives and the extension and outreach objectives outlined in this project, by focusing on supporting the production system with data analysis and interpretation, as well as getting feedback concerning the project and how the project supports the decision-making process. The team will meet monthly throughout the duration of the project, and will include the project director, the project coordinator, and the extension specialist. All other key research personnel will also be encouraged to join all meetings. To ensure maximum participation, the meetings will be held in person with the option to join remotely (i.e., via Zoom or WebEx). The project coordinator will present the detailed plans and the progress towards respective milestones to the advisory board at least twice per year. It is expected that the team will successfully engage multiple production systems in the project by actively collecting input from three prominent production systems (see letters of collaboration). Also, this will lead to the development of effective training materials, deployment of training modules, and support for web-based interactive platforms. The PROSPER producer support & communications team will also function as an effective industry-academia working team, which is crucial to align on-going and future research priorities, and enable collaborative field-based research in swine production systems-operated sites.Objective #5 (Extension and outreach): Promote the concept and disseminate applied knowledge derived from Precision Swine Management solutions.Working closely with the PROSPER producer and communications team, the advisory board, and with our extension specialists, we will implement the following sub-objectives: Training workshops at established swine conferences: At least five precision swine health and production management workshops will be organized and delivered in established conferences attended by the project's target audience. Conferences include the American Association of Swine Veterinarians Annual Meeting, Allen D. Leman Swine Conference, Pork Industry Conference, and James McKean Swine Disease Conference. Also, we will take advantage of the growing popularity of webinars and partner with the Iowa Pork Industry Center (IPIC) to reach the target audience using Zoom or WebEx platform (Iowa State University has the pro-version licenses of these).Development of training materials: Materials on precision animal agriculture and the findings from the research activities of this project (i.e., predictors of swine performance under field conditions) will be summarized in various formats including podcasts, fact sheets, webinars, popular press articles, and scientific manuscripts.