Παρουσίαση μεταπτυχιακής διπλωματικής εργασίας - Βούλγαρη Δέσποινα

Date & Time: Tuesday, 31 March 2026, 14:00

Thesis Title: Genomic Context of microRNAs: Integrating Deep Learning with Capture Long-Read Sequencing

 

Abstract: The aim of this thesis was to identify the ways by which one can effectively identify the transcription initiation site (TSS) of human miRNAs. It is known that some of the miRNAs are processed to their final form within as few as a few seconds of leaving the DNA sequence. It is extremely hard to pinpoint the exact location from which the miRNAs are being expressed. To effectively tackle this problem, a multi-tiered system was created using a deep learning model known as ”DeepTSS,” which is a multi-branch CNN.

The ”DeepTSS” model integrates FANTOM5 CAGE data from protein-coding genes with CLS (Capture Long-read Sequencing) data from the Gencode 2024 release to identify the TSS of the miRNAs. Using a high confidence level of 0.9 with a window size of 66 kb to maximize the specificity of the data, it was possible to identify the high-confidence promoters of the ”orphan” miRNAs with 45.59% (279/612). It was not only possible to identify the promoter regions of the previously unannotated miRNAs but also the fact that 70% of all known miRNA clusters could be retrieved. We also discovered that 11 shared regulatory nodes are involved in the regulation of several precursors by polycistronic transcription.

In conclusion, our research provides a more detailed view of miRNA-mediated gene regulation and the role of miRNAs in human diseases.