Deep learning methods for robust signal aggregation and spatial relationship modeling in proteomics--Multimodal integration of genomic, transcriptomic, and clinical data to enhance the diagnostic classification and risk prediction of myelodysplastic syndrome.
Adil El Yaalaoui X Vincent Fourmigue
First presentation: Spatial proteomics is a powerful approach for exploring the organization of biological tissues and the characteristics and interactions of the cells that compose them. Recent computational tools provide analysis pipelines for this technology, including modules for image segmentation and spatial analysis. In this context, our work focuses on aggregating tissue information at the single-cell level. Accurately capturing protein expression for individual cells is a challenging task that is often underestimated, resulting in expression profiles that may not fully represent the true cellular state. Our objective is to develop unsupervised models for spatial domain characterization and biomarker discovery.
Second presentation: Myelodysplastic syndromes (MDS) are blood disorders that are characterized by abnormal and inefficient production of blood cells in the bone marrow. The only curative treatment is allogeneic stem cell transplantation, though approximately one-third of patients relapse even after this intervention. These challenges motivate intensive research to improve MDS diagnosis, prognosis, and treatment. These syndromes result from acquired genetic abnormalities in hematopoietic stem cells of the bone marrow. Over time, additional genetic alterations may worsen the disease and lead to transformation into AML. While cytogenetic abnormalities and gene mutations provide valuable prognostic information, they capture only part of the disease complexity. Comprehensive transcriptomic characterization through RNA sequencing, integrated with genomic and clinical data in a well-characterized patient cohort with longitudinal follow-up, constitutes a significant research advancement in MDS, enabling unprecedented molecular profiling to identify novel disease subtypes and prognostic biomarkers, ultimately aiming to improve personalized treatment strategies and patient outcomes.