Séminaire IBS : Présentations autour du cluster MIAI

Par Dr Sophie Achard, Directrice scientifique de l’Institut multidisciplinaire en intelligence artificielle (MIAI)

MIAI Cluster is an AI institute bringing together partners from the public and private sectors and aiming to develop AI for human beings and the environment. We aim to develop frugal and trustworthy embedded, interactive and generative AI at the service of Humans and the environment. I will give a short description of the Multidisciplinary Institute in Artificial intelligence.
Thomas Burger and Pascal Mossuz will give a talk with a focus on their research partly funded by MIAI.

Présentation par Dr Thomas Burger : Statistical learning in MS-based proteomics : an overview of some results from the MIAI multiomics chair
Within the " Artificial Intelligence for high-throughput biomedical investigations" (a.k.a. multiomics) chair, several works have focused on improving the processing of mass spectrometry based proteomics data. This talk briefly presents two of them. The first one focuses on methods to impute the multiple missing MS measurements, despite their manifold origins (including unknown censorship). To do so, we propose to leverage correlation structures at the biochemical and at the multiomics levels, which can be learnt from the data. The second one focuses on multiplexing strategies that can be efficient to reduce the amount of missing values. However, as multiplexing unfortunately leads to a blind source separation problem when it comes to process the acquired data, we have proposed a new compressive dictionary learning strategies to resolve it. Finally, I’ll provide a brief overview of our current follow-up projects.

Présentation par Pr Pascal Mossuz (Service d’hématologie biologique-CHUGA & Institut pour l’Avancée des Biosciences (IAB)) : AI strategies for improving diagnosis and prognosis of Acute Myeloblastic Leukemia (AML)
Acute myeloblastic leukemia (AML) is a blood cancer which, despite advances in therapeutic intensification and allogeneic marrow transplantation, still has a poor prognosis, with an average 5-year overall survival of 35%. Against this backdrop, the identification of prognostic and predictive markers that would enable us to better assess the prognosis of patients at diagnosis, as well as to implement strategies for the early diagnosis and prediction of AML, is a major challenge for improving patient survival. The accumulation of data in public databases (e.g. TCGA) or in hospital health data warehouses enables us to develop artificial intelligence strategies on multimodal data to meet these challenges. In our team, we have developed two lines of work : i) A better diagnostic and prognostic characterization of patients by new biomarkers, identified by machine learning strategies on their initial data (Haematologica - Vol 107 june 2022 ; Lancet Digital health 6-e323-33, patentFR24/11078) ii). The search for factors predictive of the emergence of AMLs at an early stage when no biological marker is yet available to make the diagnosis. To this end, we are retrospectively investigating data collected in the CHUGA health data warehouse (Predimed) on a cohort of 350 AML patients. Our aim is firstly to characterize the medical trajectories of these patients, and secondly to apply causal inference strategies to identify clinical or biological events, which would make it possible to identify patients at risk at an early stage of the disease (IRGAMIAI 2024_Post Doc).

Hôte : Cécile Morlot (IBS/Groupe Pneumocoque)