Automating the production of animal health decision support software based on realistic epidemiological models

Context :

Mechanistic models play a major role in understanding, predicting and controlling the spread of pathogens: they allow the representation of hypotheses and processes with a high level of realism, and the comparison of a large number of scenarios, including the evaluation of control measures not yet implemented in the field. The development in the BIOEPAR unit of an open source epidemiological modelling software, EMULSION, which uses artificial intelligence (AI) methods, facilitates the writing, validation and reuse of such models. However, in order to explore the behaviour of the system, the parameterisation and outputs of academic models are more numerous and diversified than for targeted use in the field. Moving from the academic model to a decision support tool (DST) that can be manipulated autonomously by non-modellers to manage concrete health situations remains a step that requires significant software development. The ATOM project aims to automate this process by combining AI and software engineering methods.

Results :

This project (coordinated by S. Picault and P. Ezanno) started in February 2020 for 18 months. The engineer in charge of developing the necessary software chain, G. Niang, has already produced a first prototype of a web tool capable of interacting with EMULSION to control the parameterisation of a model, run simulations of various scenarios, and present the results in graphic form.

It is developing a dedicated language (DSL) that allows a DAT designer to specify the characteristics of the tool he wishes to produce from an EMULSION model of interest: free parameters, a set of parameters reflecting animal husbandry practices or control methods to be explored, scenarios to be compared (without disease, with disease without control, with vaccination or change of practice, etc.) and graphs summarising the results. This information will make it possible to automatically produce operational web pages constituting a DMO "skeleton", which can then be adapted according to the intended professional use.

This project is supported by site synergies between Oniris and IMT Atlantique, via the EpiSoft project (beginning of 2021) which will explore the implementation of software engineering methods (M. Tisi, IMT Atlantique) to facilitate and make reliable the development of epidemiological models in the EMULSION and ATOM modelling languages

Perspectives :

Accelerating the transition from realistic academic models to DMOs that can be used by animal health managers (farmers, veterinarians, technicians, animal husbandry consultants, etc.) will promote the transfer of knowledge from modelling research teams to the professional sectors and inform public decision-making.

Furthermore, the open nature of the methods developed (open source software; models that are readable in their structure, hypotheses and parameterisation; explicit description of the structure of the generated DMO) will ensure the transparency of the modelling process and open up the conclusions that can be drawn from it to discussion.

Finally, the research synergies developed during this project will make it possible to develop innovative computer methods that will find applications in many fields of modelling beyond the sole framework of epidemiology.

Principle of the ATOM project - Based on the experience gained in the production of ad hoc DMOs from academic models, the ATOM project aims to develop the software chain to automate the transformation of such models into operational DMOs, to be finalised for their professional use.