Computational methods to identify and characterize the functionality of polyadenylation isoforms

Polyadenylation is the process by which poly(A) tails are added at the 3’ end of an RNA molecule. Alternative polyadenylation (APA) can occur when multiple poly(A) sites exist. This gives rise to transcript isoforms which can have different 3’ untranslated regions (3’UTRs) or can lead to different protein products.
APA has been shown to influence the stability, localization and translation of the mRNA molecules. APA is known to be involved in health and disease and is tissue- and cell type-specific. Although a lot is known about the processing machinery, less is known about the context-specific selection of a specific poly(A) site. With the appearance of single-cell transcriptomics (scRNA-seq), it got possible to study APA on the level of individual cells.
I developed SCUREL, a computational method that detects 3’UTR changes between two sets of cells. We found SCUREL to be more sensitive compared to a similar method. Applying SCUREL to lung tumor, we found that the global 3’UTR shortening in tumor tissues cannot be explained by the proliferative cancer cells alone, but by a combination of most cell types composing the tumor tissue. Additionally, the proteins targeted by 3’UTR shortening are mostly implicated in protein metabolism and localization processes.
Since we noticed that the 3’UTR annotations were incomplete, we developed a computational method that identifies novel poly(A) sites from scRNA-seq and termed it SCINPAS. It extracts poly(A) containing reads and identifies poly(A) sites irrespective of genome annotation. We assessed the performance of SCINPAS on systems with known effects and against a competing method. We demonstrated its usability and its ability to detect novel poly(A) sites in genic and non-genic regions.
I have been involved in large collaborative projects, such as APAeval, where I was a main co-organizer. The APAeval hackathon was a community-driven effort to evaluate tools related to APA analysis based on conventional RNA-seq data. We built a benchmark suite for the reliable and reproducible assessment of the tool’s performance against ground truth data. Furthermore, I co-developed ZARP, an RNA-seq workflow, which performs the basic steps of an RNA-seq analysis in an automatic and reproducible manner. ZARP follows best programming practices.