Computational dissection of cellular distributions and cell-type specific regulation patterns from spatial transcriptomics data

Cancerous tissue not only contains cancer cells but also a variety of different immune cells. Some of them are helpful, others change sides and now assist the tumor. The knowledge of the immune cell type distribution of a patient is very important. This information helps to define the best personal treatment, give a clearer survival prognosis or to define new possible treatment options. Additionally, immune cells in the tumor microenvironment undergo several changes, some helpful in the fight against cancer, others assist the tumor to grow or spread. The knowledge concerning these changes in the different immune cell types help explain the processes in the tumor or to detect possible new biomarkers for treatment options. In this project we want to regard both the spatial distribution as well as the changes of the different immune cell types depending on their localization in relation to cancerous tissue. Which changes can we find? How do they affect each other and how are they affected by the tumor? We want to use this knowledge to detect new possible biomarkers for treatment options.

Spatial transcriptomics provides unprecedented details about the molecular mechanisms that take place within and between cells. ST assesses the molecular content of well-defined regions in a tissue which potentially cover contributions from multiple cells. In order to draw correct biological conclusions from ST analysis, spatial distributions of individual cell types are a prerequisite. This generates an urgent need for computational methods addressing this question. However, computational estimates of cellular compositions from ST are complicated by several factors. First, the methods require well curated data which is usually not available; every specimen has its own individual characteristics and so, every tissue type or even every patient, might require a own, comprehensively verified model. Second, data is noisy, making estimates prone to errors. Moreover, even if a computational model is designed for a specific tissue or even patient, it might be miss-specified, leading to additional systematic errors. In this project, we will develop a toolbox called Computational Dissection of Cellular Distributions (CDCD), which directly accounts for these issues. CDCD will advance the state of the art by (1) taking into account a priori unknown, hidden cellular background contributions, (2) taking into account environmental factors (i.e., tissue and patient dependence), and by (3) comprehensive errors estimation. Intriguingly, in contrast to available techniques, CDCD will facilitate estimates of cellular regulation in space! Such gene-regulation gradients will allow us to address questions of how immune cells adapt their molecular phenotype with respect to their environment. Additionally we will verify our algorithms on a conjoint norwegian phase II melanoma case study. CDCD will be made freely available via easy-to-use webpage and software solution, enabling both expert bioinformaticians and wet-lab biologists an efficient use of the to-be-developed methodology.