Artificial intelligence methods for modelling the spread of pathogens in highly structured populations in space and time

Context and issues

Mechanistic epidemiological modelling, which explicitly represents the processes of a pathosystem, is crucial for understanding and anticipating the transmission of pathogens in populations, and for comparing the feasibility and effectiveness of realistic control strategies. The development of such models is complex, as they must represent the dynamics of propagation, the structure of populations, interactions with their environment, etc. Artificial intelligence (AI) methods can be used to develop modelling frameworks that facilitate the expression, writing, validation and reuse of models in a form that can be read by non-modelling scientists (e.g. the EMULSION software developed at BIOEPAR). However, in certain livestock systems, the strong spatial and temporal structuring of populations can lead to complexity that must be taken into account in order to propose realistic measures. Facilitating the development of such models, while preserving the generic nature of the methods used, remained a challenge, to which the work published this year in the international AI conference ICAART responds.

Results

The article published at ICAART proposes the formalization of the couplings and constraints between the various population structures (spatial and temporal) by means of methods at the confluence of AI (multi-agent simulations) and software engineering. An original, generic and reusable abstraction has been developed and integrated into the EMULSION software, thus allowing the modelling of complex pathosystems. The associated modelling language allows to describe these structures without writing code.

As a proof of concept, this approach was applied to model the strip management of a pig farrow-to-finish farm. The representation and coupling of the different organisational levels (bands and litters; sectors, buildings and rooms) are facilitated and can be modified as required. This method makes it possible to monitor the status of individuals and groups of individuals, their evolution, and to parameterise their relationships to reflect realistic situations (number of bands, rooms, bands per room, assignment according to physiological status, etc.). These specifications are managed automatically by the simulation engine extension. The simulation results, which are in line with the actual conduct in the field, and the exchanges conducted with biologist and veterinary researchers, confirm the relevance of this approach.

Perspectives

The ability of epidemiological models to explicitly take into account the strong structuring of populations in space and time makes it possible to finely parameterise transmission routes, to identify precise control levers in relation to realistic practices, and to integrate finer grains according to the hypotheses to be explored, without impacting the structure of the model. This organisational model will be applied to the modelling of swine influenza (influenza A) and porcine reproductive and respiratory syndrome (PRRS). In the longer term, it should facilitate the integration of finer (intra-host) or broader (inter-population) scales in the same model in a homogeneous manner.

Bibliographical references

Sicard, V., Andraud, M., Picault, S., 2021. Organization as a Multi-level Design Pattern for Agent-based Simulation of Complex Systems:, in: Proceedings of the 13th International Conference on Agents and Artificial Intelligence. SCITEPRESS. doi:10.5220/0010223202320241

Organisational system applied to a farrow-to-finish pig farm, to represent social structuring in bands/litter and spatial structuring in sectors/barns.
The same mechanisms are used for both types of organisation.