This research project focuses on the detection of causal links in knowledge graphs representing multi-scale and multi-objective problems, particularly present in the life sciences.

Several issues need to be addressed. Firstly, life science ontologies involve many concepts and attributes, leading to a particularly large search space in these fields. Secondly, life science datasets may contain many imprecise and missing quantitative data, and the uncertainty of data and scientific knowledge must be taken into account. The third main concern is the interpretability of results. Particular attention will be paid in the project to helping end-users (i.e. experts in the field) to understand, evaluate and exploit the causal links detected.

These challenges will be applied in two distinct areas of the life sciences that are linked to environmental issues in plant development. Plant growth and development are tightly controlled by genotype, environmental cues and the interaction between the two (GxE). Phenotypes measured on the same genotype in different environments often show significant environmental effects, revealing phenotypic plasticity. Conversely, a robust phenotype may be considered insensitive to the environment. In a context of climate change, phenotypic plasticity or robustness can confer adaptive values on organisms. This is a major challenge given the preoccupation with global warming, which has enormous socio-economic, industrial and political impacts.

This research project aims to develop a new approach for the automatic detection of graded causal rules that express causality between different variables in knowledge graphs.

The approach developed will exploit temporal dependencies, contextual identity links and statistical methods. Targeted applications are seen as decision support systems in the following fields: (i) maize, where domain experts are interested in how climate signals impact differently on the development of different genotypes, and thus address the issue of organisms' adaptation to climate change; (ii) rice, where domain experts are interested in determining gene-gene interactions and their interactions with environments.

Contacts : Juliette Barthélemy | Fatiha Saïs