Artificial Intelligence in Drug Discovery for Health – AIDD4H-v2

AIDD4H-V2 is a LabCom in the field of personalized medicine, initiated by the Oncodesign a biopharmaceutical company and the CIAD EA 7533 laboratory, specialized in Hybrid and explainable Artificial Intelligence. AIDD4H, for Artificial Intelligence in Drug Discovery for Health, focuses more specifically on the identification and validation of new therapeutic targets. This topic is a major innovation challenge for Oncodesign, giving it a competitive advantage in the development of new therapeutic molecules on behalf of its customers and for its pipeline. The Data Revolution in the healthcare sector is generating massive amounts of heterogeneous, temporal, multi-scale and multimodal data. To understand the complexity of cancers and serious diseases without known treatment, at the heart of Oncodesign's mission, and to allow the identification of new therapeutic targets, it is now necessary to integrate, annotate and enrich this data with the know-how in the domain of drug discovery chain.

AIDD4H wishes, in combining the expertise of CIAD and Oncodesign’s DataScience Lab, to exploit the contribution of Artificial Intelligence by federating the approaches of modeling by data and modeling by knowledge and orient the associated research projects around the notions of Hybrid AI coupling Machine Learning and Knowledge Representation and Reasoning ( KRR) and Explainable AI integrating the knowledge of domain experts at the heart of the algorithm. These Explainable AIs combine connectionist AI approaches such as deep learning, neural networks ... and causal AIs based on the modeling of causal graphs of knowledge derived from the knowledge of experts. They help in the construction of explainable models, in the extraction of implicit / hidden knowledge, in the understanding of the complexity of biological entities and their interactions, in the discovery of new mechanisms of action and therefore ultimately in the identification of new therapeutic targets. AIDD4H, in addition to the complementary skills and know-how of the CIAD laboratory and Oncodesign, will also benefit from the data and knowledge generated during the PSPC project OncoSNIPE®, dedicated to the identification and characterization of patients resistant to anti-cancer treatments, and the PSPC IMODI®, dedicated to the characterization and development of predictive animal models in oncology. Explainable AI implemented by AIDD4H will make it possible to mitigate the “black box” nature of algorithms, bringing the necessary transparency to the decryption of hidden connections and complex contexts, to explain the choice of AI and help researchers and doctors in their research activities.

To solve these issues, AIDD4H is structured around three work packages:
WP-1 concerns the construction of a knowledge base by aggregating heterogeneous proprietary or public data sources. This axis will answer the question of the modeling and the acquisition of heterogeneous knowledge.
WP-2 concerns the qualification of the truth and the value of this raw data to extract the implicit / hidden knowledge. This axis will use data mining, statistical and probabilistic analysis or machine learning approaches.
WP-3 concerns the numerical and formal modeling of the reasoning involved in a know-how. This knowledge, resulting from interviews with domain experts, will be represented in a formal description logic model. Vocabulary and set of rules thus generated will allow experts to express their reasoning within developed AIs but also to AI the ability to explain its reasoning. This WP-3 will allow to combine connectionist and symbolic approaches to make these AI explicable.
The innovation resulting from this WP-3 will be integrated into a target discovery platform developed in parallel by Oncodesign, which will allow by 2023 to accelerate the research and development phases of new molecules and to create new offerss that will ensure the sustainability of the LabCom.