Mots-Clés

AI- Therapeutic Target - Machine Learning - Precision Medicine - Modelling approaches - Oncology

Description

About the position:

Biological networks are very effective tools for modeling, analyzing and discovering new biological interactions in complex biological systems. In recent years, network models and algorithms have been used to for the development of precision medicine for many diseases. The mathematical machinery at the heart of this research field is based on graph theory, a widely studied disciplinary field. This is also associated with machine learning on structured data in the form of graphs.

A big challenge is to create better modelling approaches to integrate human expertise and artificial intelligence techniques. This integration aims to limit bad effects of purely data-based methods such as:

• unforeseeable effects related to the poor quality and bias of the input data which can impact the purpose and result of the algorithm or the inability of the AI to process certain external data, to return or supplement them reliably in the absence of sufficient reference data.
• Conformist filtering and confinement effects by reducing the ability to discover and understand something else because the algorithm categorizes the data in predefined categories
• "Black box" effects: due to the opacity and lack of transparency of the AI algorithmic system. The healthcare professional and patient alike have no information and explanation on how it works AI.
• Gluttony for data. Data driven models require a lot of data to be properly trained and this can be problematic when dealing with patient data.

Our challenge is to exploit big data (more in terms of dimensionality and heteromodality rather than quantity) for clinical research and drug development, to advance better understanding of pathologies and formulate hypothesis on new mechanism of actions and identify corresponding innovative therapeutic targets. To meet this challenge, many emerging works propose the design of explainable AIs, allowing the identification of innovative therapeutic targets by drug discovery expert knowledge instillation. These explainable AIs combine connectionist AI approaches such as deep learning, neural networks, etc. and causal AIs based on modelling causal graphs of knowledge derived from the knowledge of domain experts. Such human knowledge instillation to modelling approaches lead to enhance AI explainability by constraint.

This research will address questions such as:

1. How to aggregate data sources from heterogeneous biological and medical databases while maintaining the consistency of the associated knowledge?
2. How to enhance models explainability with knowledge expert integration, based on knowledge elicitation process (bibliographic watch, expert interviews and participant observation)?
3. What are the best functions for analyzing raw data to extract knowledge?
4. How to develop in silico prediction and validation models for new therapeutic targets which will then be validated in vitro and / or in vivo?

Required selection criteria:

The qualification requirement is that you have completed a master’s degree with a strong academic background in one or more of: biology, computer science and engineering, mathematics or equivalent education with a grade of first third of the promotion. The candidate must have background in computer science with ideally skills in Machine Learning and/or Knowledge Engineering. Knowledge in the field of Biology will be required (and ideally in Oncology). Applicants must provide evidence of good English language skills, written and spoken. Mastery of the French language will be appreciated.

Preferred selection criteria:

Background in Artificial Intelligence and/or Data Mining/Data Science applied in Medicine and Biology. A candidate with some industrial experience in aforementioned areas will get preference. Publication activities in the aforementioned disciplines will be considered an advantage.

Salary and conditions:

Ph.D. candidates are remunerated and employed by the company. The appointment is for a term of 3 years and can be extended beyond the Ph.D. defense. Appointment to a Ph.D. position requires that you are admitted to the Ph.D. program in computer science and that you participate in an organized Ph.D. program during the employment period.

About the application: This research is funded by the French government and the Oncodesign SA (https://www.oncodesign.com/en/) in the frame of CIFRE Ph.D. (Convention Industrielle de Formation par la Recherche). Oncodesign and the CIAD laboratory have initiated a scientific collaboration in the field of precision medicine. This collaboration concerns the identification and validation of new therapeutic targets and the acceleration of the research and development phases of new molecules.

The job will be located at Dijon, a gastronomic and touristic French city at 1.5 hours from Paris by train. The CIAD Lab and Oncodesign are 500 meters distance.