Société Française de Bioinformatique

Mots-Clés

toxicity adverse outcome pathway AOP knowledge graphs SPARQL CYPHER

Description

The main objective of cosmetic industry is to develop products that are both safe and effective. This requires to rigorously evaluate the bioactivity of all cosmetic ingredients.
To align with evolving regulatory and ethical standards, we prioritize New Approach Methodologies (NAMs). These innovative methods combine in silico and in vitro techniques, and rely on testing strategies that take causality into account to predict how substances interact within biological systems1. For these methodologies to be effective, they must:

• Evolve with science & knowledge: ensuring relevance and accuracy.
• Promote transparency: Build trust with users and gain acceptance from regulatory authorities through clear, reproducible processes.
• Address diverse mechanisms: Apply broadly to biological pathways, whether for assessing safety or demonstrating efficacy.

The core challenge  lies in developing methods that not only accurately predict the potential toxicity of molecules but also anchor these predictions in the most current biological knowledge[1].

The objective of the internship is to optimize a testing strategy development process, based on knowledge graphs developed internally. This strategy should be able to identify all the possible sequences of biological events that may result in the toxicity of a molecule, and the optimal set of biomarkers to measure in order to ensure their exhaustive and optimal coverage of these paths. It will participate in guaranteeing the safety of products and in prioritizing the molecules to study. Such development relies on identification of nodes of interest from both statistical and biological interest, define metrics of selection. Final output being a knowledge grounded, explainable and transparent testing strategy to be proposed to toxicological experts.
To reduce complexity of systemic toxicity, two use cases, focusing on specific endpoints will be define.
The proposed approach consists of the following steps:

• Develop a knowledge graph for representing the biochemical interactions in a context of interest such as hepatotoxicity or reprotoxicity. This will require to integrate reference knowledge bases such as Reactome for the biological pathways and reactions, ChEBI for the small molecules, possibly PubChem, as well as toxicity-oriented resources such as AOPs (Adverse Outcome Pathways) and PhysiologicalMaps, and internal resources developed at L’Oréal.
• Develop an analysis method and the associated metrics to identify the important nodes and paths in the knowledge graph that connect a molecule to toxic outcomes
• Develop a knowledge grounded, explainable and transparent biomarkers testing strategy based on these important nodes and paths.

The ideal profile is a bioinformatician capable of manipulating knowledge bases and of understanding biological events, interested in developing skills in graph analysis, knowledge base reasoning, toxicology. Proficiency with Python and git is a must. Knowledge of SPARQL and/or CYPHER is a plus.
The 6 months internship will start in January or February 2027. It will be located in the BioGraphs team at IRISA (former Dyliss) and will involve some missions to L’Oreal Paris. It will be co-supervized by Emmanuelle Becker, Olivier Dameron and Anne Siegel from BioGraphs, and Kahina Abed, Léopold Carron, and Romain Grall from L’Oréal. Allowance in accordance with the statutory rate in France at the time of the internship and the rules of the host institution.

• [1] Ouedraogo et al., « Read-across and New Approach Methodologies Applied in a 10-Step Framework for Cosmetics Safety Assessment – A Case Study with Parabens »
• [2] Luechtefeld et Hartung, « Navigating the AI Frontier in Toxicology »