Unsupervised hierarchical deconvolution of gene expression data to unravel the tumor micro-environment complexity ; Déconvolution hiérarchique non supervisée appliquée aux données d'expression génique pour élucider la complexité du micro-environnement tumoral

Solid tumours are characterised by a complex organisation of the cellular ecosystem, in which the tumor cells reside and progress, called tumor microenvironment (TME). This TME is the primary target of immunotherapy that aims to critically impact tumour growth or invasive and metastatic potential. Thus, characterising a cancer patient's TME content and state becomes a priority. However, due to the large variability of the TME and its molecular complexity, it is difficult to have precise targets for such analysis in advance. For this reason, it is reasonable to use unsupervised approches that don't require any a priori knowledge. In this case, it is through the use of a matrix factorisation method named Independent Component Analysis (ICA) that we can begin to dissect expression data and extract signals related to the immune infiltration. ICA applied to expression data aims to extract independent biological signals from the data in the form of a weighted vector of genes. Although this method proved to be efficient for this deconvolution task, due to its unsupervised approach, it bears some complications when it comes to the need to select the number of signals we expect to be in the data or when we want to interpret those signals. To alleviate the problem of choosing a specific dimension for the data decomposition, a novel method was constructed that allows to screen signals over a range of multiple dimensions and retrace their evolution and behaviour through a hierarchical analysis. This approach gives not only an idea of the quality of the retried signals but also can help reconstructing relations between these signals. As for the interpretation of those signals, it is possible to use comprehensive signalling pathway reconstructions to draw conclusions on their biological meaning but also to find additional informations on a more systemic level. In definitive, it is through the use of unsupervised analyses coupled with a detailed description of molecular interactions that we can unravel the complexity of the ...