PS7-15. National Animal Nutrition Program (NANP): Harmonized Databases, Reproducible Modeling, and Artificial Intelligence-enabled Knowledge Synthesis for Animal Nutrition Research.
Mingyung Lee, Jordan M Adams, Julia Travassos da Silva, Heidi Rossow, Todd R R Callaway, Peter R Ferket, Timothy J J Hackmann, Mark D Hanigan, Hector M Menendez, Edgar O O Oviedo-Rondon, Aline Remus, Michael J VandeHaar, Sarah H White-Springer, Luis Orlindo TedeschiAbstract
Livestock and companion animal nutrition research increasingly depends on harmonized datasets and reproducible modeling workflows to synthesize evidence across feeds, genetics, environments, and management practices. The National Animal Nutrition Program (NANP; NRSP-9) supports this need for beef, dairy, swine, poultry, and horse systems by providing curated, research-based databases, modeling resources, and training that enable model development, parameterization, evaluation, and benchmarking. Core assets include a feed ingredient composition database with >4 million records and an animal performance and metabolism database with >500,000 observations; both are operational and publicly accessible. Recent infrastructure modernization includes website restructuring, Shiny-based decision tools, evaluation of alternative hosting platforms to reduce long-term costs, and hybrid R/Python workflows to enhance interoperability, provenance tracking, and end-to-end reproducibility. To strengthen transparency and corrective software stewardship, the NANP Modeling Committee established a formal process to identify, document, and route technical issues in widely used national nutrition modeling resources to original developers for recompilation and version correction, reinforcing accountability, consistent versioning, and reproducible benchmarking across releases. The latest strategic advancement is the development of a NANP Large Language Model (LLM) system using retrieval-augmented generation (RAG) to support structured knowledge synthesis while respecting copyright and licensing constraints. Deployment options include public GPU-hosted servers and stand-alone local installations to balance cost, accessibility, and data governance. These initiatives align with priorities in Bayesian modeling, artificial intelligence (AI)-enabled decision science, digital twins, and sustainability analytics, including methane accounting and integration with feed life-cycle assessment. Training and workforce development remain central, with modular modeling education, symposium integration, and certification concepts under development. Collectively, NANP strengthens national modeling infrastructure, improves reproducibility, supports corrective software governance, and advances AI-integrated decision support for animal nutrition research and practice.