Back to search

University of Thessaly (UTH)

Μηχανική μάθηση για την ανάλυση μεγάλων βιολογικών δεδομένων

Abstract

dc:description

This doctoral dissertation is part of the research areas of Bioinformatics and Big Data Analysis with Machine Learning. The primary goal is to develop computer models capable of extracting useful information from the processing of biological data, which will lead to the expansion of knowledge on open biological issues. The research focuses on two biological objectives, the functional characterization of small open reading frames (sORFs) and the regulation of coding and non-coding gene expression. Recently, an increasing variety of peptides of less than 100 codons have been identified in various organisms, from bacteria to humans, which act as regulators of many critical processes such as metabolism, growth, and cell death. In addition to the sequences encoding large proteins, there are many small open reading frames that are processed by the cell translation mechanism to produce small peptides. Small peptide chains are usually not recorded in proteomic annotations as due to their small size they are considered to occur randomly and fail to achieve high scores on functionality indicators such as conservation. The open issue of the possibility that small open reading frames encoding peptides, was approached in the present dissertation with the development of D-sORF, a machine learning Framework that integrates statistical nucleotide context and motif information around the start codon to find exact solutions of coding sORFs. The algorithm scores for coding identity directly similar to established protein coding ORFs by only requiring the underlying genomic sequence, without incorporating parameters like conservation that in the case of sORFs may increase dispersion of scores. The next biological issue in this dissertation is the characterization of promoters of coding and non-coding RNAs. CAGE protocol has emerged as a highly efficient experimental technique in identifying transcription start sites and consequently promoters. Despite its increasing popularity as an experimental promoter recognition protocol, CAGE specialization in identifying transcription initiation events in the genome has several limitations. There is strong evidence that in addition to promoter regions, CAGE recognizes capping sites along various locations of transcribed loci such as different splicing products, isoforms and capped molecules that can be summed up as transcriptional noise. As a result, only a subset of the CAGE-enriched regions was found to overlap with the surrounding area of annotated TSS. This is a major obstacle in studies aimed to integrate regulatory regions into the framework of biological pathways.Aiming to distinguish the actual transcription events we developed ADAPT-CAGE, a high-resolution algorithmic framework that use a multilevel Machine Learning model. It utilizes structural features of the DNA sequence, POL II binding motifs and gene expression for scoring at each CAGE peak. In addition to the machine learning model, a multifaceted evaluation framework based on both experimental data and computational methods was developed. Comparison with corresponding CAGE data processing algorithms for TSS identification in all individual evaluations showed quality of the ADAPT-CAGE results. The ADAPT CAGE algorithm was then applied to a multitude of samples from cell lines, primary cells and tissues to identify miRNA promoters. The identification of miRNA promoters is a fundamental effort to understand and characterize the underlying mechanisms of both physiological and pathological conditions. Detailed annotation of transcription start sites -TSS and consequently promoters, combined with experimentally and computationally identified transcription factor binding sites for each sample, entered a repository of miRNAs regulation data. The repository is called miRGen v4 and is publicly available to the scientific community. The Dis-TSS algorithm was a continuation of the research in the field of detecting real TSS events from CAGE signals. In this effort signal processing techniques were combined with Machine Learning. Features from the spatial and frequency domain were extracted from the CAGE peaks and trained a machine learning system to separate transcription initiation events from noise. The significance of each feature was assessed and only those with high prognostic capability were retained. The development of reliable algorithms and repositories in the present dissertation could play a fundamental role in revealing biological conclusions by pushing the horizon of scientific knowledge even further.

Degree

thesis:*
Grantor dc:publisher
University of Thessaly (UTH)
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Perdikopanis, Nikolaos
  • Περδικοπάνης, Νικόλαος

Subjects

dc:subject × 19

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/48802

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Perdikopanis, Nikolaos; Περδικοπάνης, Νικόλαος. Μηχανική μάθηση για την ανάλυση μεγάλων βιολογικών δεδομένων. University of Thessaly (UTH), 2021. http://hdl.handle.net/10442/hedi/48802