University of Crete (UOC)
Further development of UniReD tool for the prediction of protein protein interactions to assess carotid artery disease and major depressive disorder comorbidity
Abstract
dc:descriptionProtein-protein interactions are a central research focus for understanding complex biological systems. Large-scale experimental methods developed in recent years for identifying protein-protein interactions enable the simultaneous investigation of numerous proteins. However, these methods face significant limitations related to the accuracy of results and the cost in resources. Complementary to experimental approaches, computational methods exist for predicting, recognizing, and evaluating both direct and indirect protein-protein interactions. The computational approach uses data on protein structures (genomic and proteomic), information derived from phylogenetic analysis, gene neighborhood, and gene fusion events to predict interactions among proteins under investigation. Another computational technique involves knowledge extraction from the literature, primarily using abstracts from published articles, proposing potential protein interactions based on metrics computed from their “co-occurrence” in documents. Numerous tools have been developed to extract information from PubMed articles for identifying and predicting protein-protein interactions. Additionally, databases record interactions from large- scale published experiments, many of them using standardized record curation systems. Computational tools combining all available sources aim to consolidate this information. A computational tool for predicting protein-protein interactions is UniReD. Starting with all articles describing and supporting each protein's function in the UniProt database, it collects all related articles from PubMed to them. These articles are then clustered using a machine-learning algorithm. The articles are then replaced by their associated proteins to create clusters of proteins likely to be related. This doctoral dissertation explored an expansion of UniReD's functionalities and applying it to investigate comorbidities of diseases, focusing on carotid artery disease and depression. The improvements targeted three main directions: including available information beyond the literature to rank predicted interactions, incorporating experimental data to enhance ranking, and comparing protein interactions to identify shared biological mechanisms in two diseases. In line with other computational tools that integrate information from diverse sources, methods and algorithms were developed to rank interactions predicted using UniReD's methodology. Information on proteins predicted by UniReD was collected. This included data on Homo sapiens and Mus musculus proteins and their ontology. At the genetic level, paralogs of each gene from Ensembl and orthologous genes between H. sapiens and M. musculus were recorded. Protein complex information and participants in each complex were gathered from ComplexPortal, while functional domains for each protein were obtained from InterPro. To determine whether any H. sapiens proteins appear as fused in M. musculus and vice versa, searches were conducted using the DifFuse and CompositeSearch algorithms on all proteins from both organisms. Protein sequences for H. sapiens and M. musculus were retrieved from UniProt, and sequence similarity searches using BLAST (blastp) were conducted within and between the organisms. The data were structured into relational databases to enable protein-protein interaction ranking predicted by UniReD. For instance, a functional relationship for a pair of proteins predicted in H. sapiens could be determined if these proteins are fused in M. musculus or if their orthologs interactin M. musculus. Similarly, the interaction could be supported by whether paralog genes, protein complexes, or protein clusters predicted by UniReD confirm the relationship. Using this data, a scoring scheme was developed, and a graphical interface called multiple UniReD was created, enabling users to input two protein lists and rank them according to this scheme. The tool was used to rank methylation biomarkers against proteins known for their roles in breast cancer, osteoarthritis, and diabetes melitus. 4Another goal was to include large-scale experimental data to further support UniReD's predicted interactions. Given the rise of single-cell RNA sequencing (scRNA-seq), a tool was developed to analyze such data and identify gene correlations, named scRNA-Explorer. The tool allowed for quality assessment of sequencing data and cell exclusion based on user-defined parameters. After grouping cells, the expression of key biomarkers for each cell type was examined. Correlation analysis could then identify correlated genes to a “bait” gene and subsequently look for enriched biological processes and pathways of the correlated genes. Results could further refine the ranking of UniReD’s predictions. In the search for comorbidities, methods were developed to link genes/proteins to diseases using the literature. Two scoring formulas were tested, relying on a database of PubMed articles. Using these, proteins co-referenced with the two diseases under investigation were retrieved. Relationships among these proteins were examined using the scoring scheme developed in multiple UniReD. Proteins unrelated through any mechanism (paralogs, orthologs, complexes, fused proteins, UniReD clusters) were excluded. Ninety-seven shared proteins were identified, and Reactome pathway enrichment analysis revealed their involvement in inflammation-related processes through cytokine and interleukin activity. This finding aligns with modern evidence linking inflammation to cardiovascular disease and mental health. Further analysis with proteins involved to these pathways could highlight other candidate proteins and biological processes. In conclusion, UniReD, with its enhanced capabilities, provides a robust tool with the potential to assist experimental design and prioritization of potential biomarkers for experimental validation.
Degree
thesis:*- Grantor dc:publisher
- University of Crete (UOC)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Baltsavia, Ismini
- Μπαλτσαβιά, Ισμήνη
Subjects
dc:subject × 14- Πρωτεϊνικές αλληλεπιδράσεις
- Συσχέτιση έκφρασης γονιδίων
- Εξόρυξη γνώσης
- Protein-protein interactions
- Text mining
- scRNAseq
- Metabolic pathways
- Gene-gene correlation
- Βιολογία
- Φυσικές Επιστήμες
- Μαθηματική και Υπολογιστική βιολογία
- Biological Sciences
- Natural Sciences
- Mathematical and Computational Biology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58289