Speakers
Description
Animal disease surveillance generates large volumes of heterogeneous data. The current challenge is to integrate them in a timely manner, with traceability and quality assurance, so that they can be translated into actionable alerts. Here we present AI4AI, an early-stage pilot project that aims to apply artificial intelligence (AI) models to epidemiological, genomic and environmental data and link them to operational alert systems.
Highly pathogenic avian influenza (HPAI) is the first use case. Its sustained circulation in wild birds, its spread to domestic and wild mammals, and its zoonotic potential call for surveillance that combines computational scalability with practical application.
The project will advance through three interrelated work packages: 1) A platform will be developed to integrate, curate and assess the quality of data from multiple sources; 2) AI will be incorporated into existing genomic tools to support the identification of lineages, mutations, evolutionary patterns and potential signals of host adaptation; and 3) DiFLUsion, an HPAI early-warning system operational since 2021 and used by Spain's Ministry of Agriculture, Fisheries and Food (MAPA), will be enhanced with temporal and spatial models to improve risk detection and progressively incorporate new modules following validation.
A REST API connection to WOAH-WAHIS has been implemented as the first functional component under an agreement between INIA-CSIC and the World Organisation for Animal Health (WOAH). It automates the weekly download and updates outbreak notifications for notifiable animal diseases. This shared data layer provides official, traceable and reusable data to feed different models and dashboards.
Previous exploratory work by the Animal Health Research Centre (CISA-INIA-CSIC) and the Artificial Intelligence Research Institute (IIIA-CSIC) has combined historical outbreaks, millions of bird observations, migratory movements and environmental variables using supervised and unsupervised learning. Models evaluated include Random Forest, XGBoost, clustering techniques and neural networks, including graph neural networks to represent ecological and spatial connections between regions. The project will bring these separate analyses into an integrated, reproducible and scalable workflow. The consortium combines these centres' expertise with that of the Molecular Biology Centre Severo Ochoa (CBM-CSIC), which will develop the genomic component, and the Doñana Biological Station (EBD-CSIC), which will contribute wild bird surveillance and ecological knowledge, including interspecies interactions and risks to vulnerable wildlife. CESGA's infrastructure will support data curation, model training and computationally intensive workflows.
Ultimately, this project aims to connect global surveillance with risk assessment through a modular and scalable architecture. Following validation, new modules could be progressively incorporated into MAPA's surveillance environment. Avian influenza will be the first demonstrator of a reusable infrastructure applicable to other animal diseases of relevance under the One Health approach. This work illustrates how digital infrastructure can connect data, analysis and decision-making, bringing computational research closer to sustainable services for animal and public health.