Evaluating Directional and Association Methods on Single-cell RNA Sequencing Data

This thesis aims to compare and evaluate directional and association methods performance on single-cell RNA sequencing (scRNA-seq) data. The scRNA-seq enables one to study biology at a single cell resolution. Although this process of RNA sequencing opens up new possibilities, the data can be subject to technical distortions, such as a dropout where the loss of information ranges from 30 to 90 %. Thus methods that work reliably for the bulk RNA data sets may perform close to random guessing for the scRNA-seq. Therefore I present a comparison of multiple methods on both the simulated and the real data sets. The directional and nondirectional studies are separated for tests using the simulated data to prevent influencing the results by methods that detect the inference inaccurately in only one of these studies. The best performing method is then used to discover new association patterns across transcriptome and proteome in acute leukaemia cells. Secondly, I demonstrate the impact of data normalisation for association methods. Four current normalisation methods and a new approach proposed here are compared on real data. The functions are tested for a new artefact creation and the original artefact destruction. Examples of these pattern transformations are provided for each approach. The findings in this thesis suggest that the normalisation of the scRNA-seq data must be carefully handled to avoid introducing undesirable artefacts into the studying of relationships between genes.