PhD position – Explainable AI for discovering rare crop phenotypes from heterogeneous scientific knowledge

INRAE; France; On-site; EUR 21,000 - 25,000; Full time

Job summary

INRAE invites applications for a PhD position in Montpellier, France, at the interface of artificial intelligence and agroecology. The project focuses on developing explainable AI methods to discover rare crop phenotypes and assess their credibility, integrating literature and knowledge bases.

Supervisors include INRAE LEPSE, LIRMM, and GAIA Doctoral School. The PhD will be hosted within an interdisciplinary team, exploring knowledge extraction, anomaly detection, and uncertainty reasoning to

Qualifications

• Master's degree or equivalent in computer science, artificial intelligence, data science, bioinformatics, computational biology or related quantitative field.
• Experience or strong interest in topics such as ML, NLP, knowledge graphs, explainable AI, uncertainty reasoning, data integration and computational argumentation.

Responsibilities

• Conduct AI research at the interface of AI and biology to develop explainable methods for rare phenotype discovery.
• Contribute to the CropTraits project by extracting knowledge from literature and harmonizing heterogeneous data sources.

Skills

Machine learning

Data analysis

Large Language Models

Knowledge representation

Explainable AI

Reasoning under uncertainty

Computational argumentation

Scientific data integration

Python

Education

Master's degree in CS/AI/Data science

Tools

Python

Job description

Organisation/Company INRAE Department AgroEcoSystem Research Field Computer science » Other Biological sciences Agricultural sciences Researcher Profile First Stage Researcher (R1) Positions PhD Positions Application Deadline 31 Oct 2026 - 12:00 (Europe/Paris) Country France Type of Contract Temporary Job Status Full-time Hours Per Week 35 Offer Starting Date 1 Nov 2026 Is the job funded through the EU Research Framework Programme? Not funded by a EU programme Is the Job related to staff position within a Research Infrastructure? No

Offer Description

Research context and objectives

Agriculture needs to diversify in order to adapt to climate change and other global environmental challenges. Among the diversity of cultivated plants, some species or varieties may display unusual combinations of functional traits associated with original ecological strategies and potentially valuable agronomic properties. Identifying these rare phenotypes could therefore reveal new opportunities for crop diversification.

However, plant phenotypic information is currently scattered across thousands of scientific articles, databases and technical resources. Integrating these heterogeneous sources creates a major challenge: an unusual observation may represent either a genuine biological singularity or an artefact resulting from measurement, annotation, extraction or data-integration errors.

This PhD project lies at the interface betweenartificial intelligence, functional ecology and agroecology. Its main objective is to develop explainable AI methods capable ofdiscovering rare crop phenotypes and assessing their scientific credibility.

The project will address three closely connected research challenges.

1. Building knowledge from heterogeneous scientific sources

The PhD candidate will contribute to the development ofCropTraits, a global database of functional traits of cultivated plants. Large Language Models (LLMs), scientific text mining and knowledge representation approaches will be investigated to extract information from the scientific literature, harmonise heterogeneous concepts and variables, connect information from multiple sources, and explicitly represent uncertainty.

The resulting knowledge base will provide a large-scale representation of crop phenotypic diversity.

2. Detecting rare phenotypes in multidimensional trait spaces

Machine-learning approaches for anomaly and rare-event detection will be developed to explore high-dimensional phenotypic spaces and identify observations that depart from dominant crop functional strategies.
A central methodological challenge will be to move beyond conventional outlier detection: unusual observations will not automatically be considered errors, but rathercandidate biological discoverieswhose credibility needs to be evaluated.

The project will investigate methods combining anomaly scores, uncertainty estimates, similarity structures and biological context to identify potentially rare functional strategies.

3. Explaining whether a rare observation represents a discovery or an error

For each candidate rare phenotype, several competing explanations may exist: genuine biological singularity, measurement error, data-integration error or context-dependent response.

LLMs will be used to retrieve relevant evidence from the scientific literature, associated metadata and the knowledge base. Because this evidence may be incomplete, uncertain or contradictory, the project will developcomputational argumentation and reasoning-under-uncertainty approachesto formally compare competing hypotheses.

The aim is not only to classify observations as credible or unreliable, but also to provide a transparent explanation of the reasoning and evidence supporting the conclusion.

Scientific environment

The PhD will be carried out in Montpellier, France, within a highly interdisciplinary research environment involving three complementary research groups:

• LEPSE – INRAE, working on plant ecophysiology, crop diversity and adaptation to environmental constraints;
• LIRMM – CNRS / Université de Montpellier, with expertise in artificial intelligence, knowledge representation, computational argumentation and reasoning with conflicting information.

The PhD will be supervised byDenis Vile (INRAE, LEPSE), Madalina Croitoru (Université de Montpellier, LIRMM) and Lucie Mahaut (INRAE, CEFE)and will be affiliated with theGAIA Doctoral School.

This interdisciplinary setting will allow the candidate to develop methodological advances in AI while addressing an original scientific problem with direct relevance to crop diversification and agroecological transitions.

Expected outcomes

The PhD is expected to contribute both toartificial intelligence methodologyand to the understanding ofcrop functional diversity. It will develop an original framework combining knowledge extraction, rare-event detection and explainable reasoning to distinguish genuine scientific discoveries from errors in massive heterogeneous datasets.

The approaches developed in this project may ultimately be transferable well beyond crop science to other biological, environmental or biomedical knowledge bases facing the same fundamental challenge:how can we distinguish a rare discovery from an error?

Requirements

Research Field Computer science » Other Education Level Master Degree or equivalent

Skills/Qualifications

Candidate profile

We are looking for a highly motivated candidate interested in working at the interface between artificial intelligence and biological sciences.

Applicants should hold, or be about to obtain, a Master's degree or equivalent incomputer science, artificial intelligence, data science, bioinformatics, computational biology, or a related quantitative field.

Experience or strong interest in several of the following topics would be particularly valuable:

• machine learning and data analysis;
• Large Language Models and Natural Language Processing;
• knowledge representation and knowledge graphs;
• explainable AI and reasoning under uncertainty;
• computational argumentation;
• scientific data integration;
• Python and associated data-science / machine-learning tools.

Candidates from plant science, ecology or agronomy with a strong quantitative and computational background may also be considered.

Beyond technical skills, we are looking for a candidate with strong scientific curiosity, an interest in interdisciplinary research, the ability to critically analyse complex information, and good written and oral communication skills in English.