Single-cell omics and multi-omics technologies have enabled the study of cellular heterogeneity with unprecedented resolution and the discovery of new cell types. The core of identifying heterogeneous cell types, both existing and novel ones, relies on efficient computational approaches, including especially cluster analysis. Additionally, gene regulatory network analysis and various integrative approaches are needed to combine data across studies and different multi-omics layers. This thesis comprehensively compared Bayesian clustering models for single-cell RNAsequencing (scRNA-seq) data and selected integrative approaches were used to study the cell-type specific gene regulation of uterus. Additionally, single-cell multi-omics data integration approaches for cell heterogeneity analysis were investigated.

Article I investigated analytical approaches for cluster analysis in scRNA-seq data, particularly, latent Dirichlet allocation (LDA) and hierarchical Dirichlet process (HDP) models. The comparison of LDA and HDP together with the existing state-of-art methods revealed that topic modeling-based models can be useful in scRNA-seq cluster analysis. Evaluation of the cluster qualities for LDA and HDP with intrinsic and extrinsic cluster quality metrics indicated that the clustering performance of these methods is dataset dependent.

Article II and Article III focused on cell-type specific integrative analysis of uterine or decidual stromal (dS) and natural killer (dNK) cells that are important for successful pregnancy. Article II integrated the existing preeclampsia RNA-seq studies of the decidua together with recent scRNA-seq datasets in order to investigate cell-type-specific contributions of early onset preeclampsia (EOP) and late onset preeclampsia (LOP). It was discovered that the dS marker genes were enriched for LOP downregulated genes and the dNK marker genes were enriched for upregulated EOP genes. Article III presented a gene regulatory network analysis for the subpopulations of dS and dNK cells. This study identified novel subpopulation specific transcription factors that promote decidualization of stromal cells and dNK mediated maternal immunotolerance.

In Article IV, different strategies and methodological frameworks for data integration in single-cell multi-omics data analysis were reviewed in detail. Data integration methods were grouped into early, late and intermediate data integration strategies. The specific stage and order of data integration can have substantial effect on the results of the integrative analysis. The central details of the approaches were presented, and potential future directions were discussed.