Dissertation committee:
Αθανάσιος Τζιούφας, Καθηγητής, Ιατρική Σχολή, ΕΚΠΑ
Δημήτριος Φωτιάδης, Καθηγητής, Τμήμα Μηχανικών Επιστήμης Υλικών, Πανεπιστήμιο Ιωαννίνων
Βασίλειος Αϊδίνης, Ερευνητής Α’, ΕΚΕΒΕ ‘Αλέξανδρος Φλέμιγκ’
Ευάγγελος Ανδρεάκος, Ερευνητής Α’, Διευθυντής του Κέντρου Κλινικής, Πειραματικής Χειρουργικής & Μεταφραστικής Έρευνας, ΙΙΒΕΑΑ
Αντώνιος Σακελλάριος, Επίκουρος Καθηγητής, Τμήμα Μηχανολόγων & Αεροναυπηγών Μηχανικών, Πανεπιστήμιο Πατρών
Ανδρέας Γουλές, Επίκουρος Καθηγητής, Ιατρική Σχολή, ΕΚΠΑ
Ιωάννα Γαλάνη, Ερευνήτρια Γ', ΙΙΒΕΑΑ
Summary:
It is currently estimated that human species can be affected by more than 10,000 distinct rare diseases. Although each one concerns a small percentage of the world’s population, together they can create a huge economic burden. Among this long list of diseases lies idiopathic pulmonary fibrosis, the most lethal of the idiopathic interstitial pneumonias. Until now there is no curative option other than the scarcely available lung transplantation opportunities. Thus, patients are treated with anti-fibrotic agents trading time for often severe side effects. Over the next decades, disease burden is estimated to increase due to a pattern of increasing prevalence and a bidirectional relationship to COVID-19, necessitating research intensification.
In this thesis, in silico methods have been utilized in order to propose novel disease biomarkers and curative options for pulmonary fibrosis. Towards that goal we have created Fibromine, a database of manually curated and consistently re-analyzed omics data, spanning two species and numerous experimental setups. Its contents are freely accessible via the homonym online application which can also be used for real-time data combination, statistical and visual exploration. Subsequently, Fibromine-proposed deregulated features between pulmonary fibrosis and control conditions were used for machine learning prioritization of disease biomarkers. A short and a long target lists were obtained from SHAP-based explanation of the models, containing both well established and interesting novel targets. Both gene lists were capable of separating pathological from steady state samples with an at least equal performance to that of previously proposed feature sets. Overall, computational analysis has proven capable of identifying new promising disease targets once applied on high quality data. Centralization of the latter is necessary for future studies acceleration, hypothesis formation and/or validation.
Moreover, various omics data have been processed for the investigation of pulmonary fibrosis-related molecules and conditions. Data analysis from control, hepatocellular adenocarcinoma, prostate, lung and breast cancer samples described a regulatory link between DNA methylation and ENPP2 expression. Deregulated during cancer and measurable through non-invasive procedures, this regulatory connection was found of some prognostic value urging for more studies on the subject. Subsequently, microbiome exploration from murine gut, liver and lung suggested the existence of an axis connecting the three organs. Diet-induced obesity was proposed to cause dysbiosis, thus potentially impairing homeostatic balance. Next, investigation of the up-regulated lipocalin-2 expression during pulmonary fibrosis suggested the molecule in question as a biomarker for lung inflammation and respiratory functional status, motivating for further studies. In fibroblasts was found to be crucial for pulmonary fibrosis and extracellular matrix invasion. Importantly, targeting Tks5 successfully attenuated pulmonary fibrosis via down-regulating podosomes formation, thus arising as a new promising therapeutic alternative. In another report, MAP3K8 was revealed to have an anti-fibrotic character via mediating inflammation-related processes and Cox-2-mediated PGE2 production. Furthermore, examination of COVID-19 data suggested the pathologic nature of ATX/LPA axis and its importance for dendritic cells homeostasis which is disrupted during SARS-CoV-2 infection. Last, a detailed study recorded the responses of HKC-8 cells post distinct treatments with 176 stimulants. Among them, LPA acted in a pro-inflammatory fashion indicating a possible pathogenic role in chronic kidney disease.
In total, the above projects created a central high quality database of pulmonary fibrosis omics data, proposed new target genes for the treatment of pulmonary fibrosis and have also delved deeper into obscure areas of the aforementioned and similar pathologies.
Keywords:
Idiopathic Pulmonary Fibrosis, Omics data, Integration, Biomarker, Machine learning
