One of the challenges of digital biology is to develop, compare and improve learning methods, adapting them to integrate multi-source data (omics, sensors, environment, data from participatory approaches). The modelling of biological and physiological processes to develop predictive approaches and the simulation of complex biological systems are also major challenges. In particular, the robustness of the models built in axis 1 must be tested by subjecting them to fluctuating conditions (internal or external).

This research includes predictive approaches based on modelling, data assimilation and the coupling of "data-driven" and "concept-driven" approaches, taking into account the essential dimensions of prediction, i.e. planning, quantification of uncertainties and evaluation of forecasts.

Research themes

• Genomic prediction;
• Taking account of all transmissible factors in predicting an individual’s potential to transmit characteristics to its descendants;
• Application of these methods to identify bio-markers for an organism’s state or potential (resistance or stress levels, etc.), and  genetic targets to optimize networks;
• Predicting the impacts of complex chemical and biological exposure scenarios on physiology;
• Mobilizing information on states and knowledge of processes, at organism (and sub-organism) levels, to predict the functioning and trajectories of populations at different spatial and/or temporal scales.

Methodological challenges

• Select the learning methods best suited to a given problem, determining their contributions and limitations for detecting complex structures or weak signals in the data;
• Build and simulate integrative models with good predictive capabilities, allowing the effects of confounding factors in the data to be identified;
• Access unobservable quantities through model inversion (e.g. trait value estimation);
• Improve the quality and robustness of predictions (underlying issues: planning and sampling, uncertainty assessment, taking account of changing environments).