Reflection group on hybrid models for anaerobic digestion: integration of omics data

Context and key challenges

Anaerobic digestion is an environmental technology that enables the transformation of biowaste into methane-rich biogas, a renewable energy source. In the context of ecological transition, it provides an essential link in the circular economy.

Modelling, particularly in the form of digital twins, makes it possible for anaerobic bioreactor operators to anticipate disruptions to production and optimize how they manage the process. The standard mechanistic model (ADM1) is nevertheless limited in its capacity to anticipate disruptions and manage conditions to prevent their occurrence. The integration of meta-omics data and AI have the potential to improve this model by combining mechanistic knowledge with machine learning.

The GREMLHINS exploratory project, which brings together INRAE’s PROSE, TBI and LBE teams, proposes to:

• create an interdisciplinary working group to develop hybrid models integrating meta-omics data;
• develop a pilot model using available biological samples;
• organize training workshops in 4.0 technologies for the DIGIT-BIO Metaprogramme community and promote the use of these tools.

Goals and methodology

L’objectif du projet est de développer un modèle hybride qui associe le modèle mécaniste de The goal of the project is to develop a hybrid model that combines the reference mechanistic model, ADM1, used for the simulation of anaerobic digesters ⁽¹⁾, with machine learning algorithms, enabling the integration of biological (meta-omics) data, and improving predictive capabilities in methane production. The combined model is intended, in particular, to improve the representation of the stress phases (inhibition, overload, crash and recovery), which are currently either poorly modelled or not modelled at all by the ADM1 model.

The project is therefore divided into three parts:

1. Building a dataset using datasets provided by PROSE (a project partner) and produced by various previous ANR projects (incubation in continuous or semi-continuous reactors, subjected to different types of abiotic stress). At the beginning of the project, the dataset will be supplemented by DNA shotgun sequencing data.
2. Identifying the limits of the ADM1 model and correlation with biological data. The ADM1 model will be calibrated using the above datasets, to identify under what conditions the model performs either well or poorly.
3. Developing the hybrid model. This second modelling stage will aim to develop the hybrid model, using the results from the previous stage and the outcomes of the reflection workshop that is planned as part of the project.

The pilot hybrid model developed through GREMLHINS should improve the predictive capabilities of anaerobic methane production systems and will therefore be of interest to both the scientific community and operators. This model will also serve as a first step in the creation of digital twins, opening up many future project pathways. Last, the project will demonstrate the use of 4.0 tools to model a complex microbial system and will make its resources freely available to the community.

Contacts:

Project participants

INRAE units