University of Thessaly (UTH)
Algorithms for the chracterization of the microbiome and the indentification of its functional effects on the host
Abstract
dc:descriptionBacterial small RNAs (sRNAs) are key regulators of gene expression, particularly in stress response, pathogenesis, and adaptation to environmental changes. Their ability to modulate messenger RNA (mRNA) targets through base-pairing interactions has established them as central players in post-transcriptional regulation across a diverse set prokaryotic systems. Despite rapid advances in RNA sequencing technologies and experimental methodologies such as RIL-seq, MAPS, and CLASH, the comprehensive identification and functional annotation of sRNA–mRNA interactions remain limited by data fragmentation and computational challenges. This thesis presents a two-pronged approach to enhance the understanding and prediction of bacterial sRNA–mRNA interactions. First, Agnodice is introduced, a curated and publicly accessible database that compiles over 39,000 experimentally supported sRNA–RNA interactionsfrom 45 bacterial species. Agnodice integrates data from both low- and high-throughput studies, with rich metadata and cross-referencing to external genomic resources, enabling advanced querying, visualization, and data exploration. It stands as the most extensive collection of its kind and provides an essential tool for researchers investigating RNA-mediated regulation. Building upon the Agnodice resource, a machine learning (ML) framework was developed to predict sRNA–mRNA interactions. Leveraging a curated dataset of over 30,000 interaction pairs, comprising ~10,000 positives and ~20,000 negatives, allowed the compilation of a set of biological and computational features. These included binding energy, site accessibility, secondary structure, and full-length sequences of sRNAs and extended mRNA regions flanking the start codon. Through systematic evaluation of multiple ML algorithms, such as XGBoost, Random Forest, and neural network architectures, it was shown that models trained on sRNA and mRNA sequence features alone performed comparably or better than models using expanded feature sets. This suggests that raw sequence data may encode interaction-relevant patterns not captured by conventional thermodynamic models. Benchmarking of the newfound model against state-of-the-art prediction tools such as TargetRNA3 and IntaRNA2 demonstrated improved predictive accuracy and a higher proportion of correctly ranked interactions. These findings reinforce the value of full-sequence-based feature representation and support the hypothesis that sequence-level information may hold underutilized predictive potential in RNA interaction modeling. Collectively, the contributions of this thesis, through Agnodice and the developed ML framework, advance the computational study of bacterial post-transcriptional regulation. They also lay a foundation for future integrative approaches that combine experimental data, sequence information, and machine learning to unravel complex regulatory networks in microbial systems.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kotsira, Vasiliki
- Κώτσιρα, Βασιλική
Subjects
dc:subject × 18- Μικρά μη κωδικά RNAs
- Βακτήρια
- Μετά - μεταγραφική ρύθμιση
- Μηχανική μάθηση
- Πρόβλεψη στόχων
- Μεταγονιδιωματική
- Non-coding RNAs (ncRNAs)
- Bacteria
- Post-transcriptional regulation of gene expression
- Machine learning
- Target prediction
- Metagenomics
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Βιοπληροφορική
- Computer and Information Sciences
- Natural Sciences
- Bioinformatics
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58975