National Technical University of Athens (NTUA)
Development of analytical frameworks for the integration of molecular data views to understand the biology of myeloid malignancies
Abstract
dc:descriptionMyeloid malignancies consist of a heterogeneous spectrum of clonal stem cell disorders driven by genetic alterations, resulting in dysregulated hematopoiesis. The investigation of the mechanisms underpinning myeloid neoplasia relies primarily on experimental models of disease biology and the phenotyping of primary patient samples using emerging genomic technologies. In recognition of the increasing complexity, scale and dimensionality of the datasets generated by these approaches, this thesis focuses on the development of analytical frameworks that operate within and across different omics modalities (genomics, transcriptomics, epigenomics) and sequencing techniques (bulk and single cell), and set out to enhance the understanding of the underlying biology of myeloid neoplasms. Specifically, this work deploys principles from multi-view data fusion and interconnection to analyze signals in an integrative manner, aiming to elucidate molecular landscapes and assist the study of phenotypes at a genetic level. Chapter 2 investigates the transcriptional repertoire and chromatin profile of SF3B1-mutated Myelodysplastic Syndromes (MDS), leveraging bulk RNA and ATAC sequencing data from patient-derived genetically matched normal and SF3B1-mutated induced pluripotent stem cell (iPSC) lines. We introduce a multi-stage fusion framework that merges signals from diverse data layers obtained from transcriptome sequencing (splicing, transcript usage, gene expression). The analytical framework developed as part of this work leads to the derivation of a splicing signature linked to 34 genes, which associates with the SF3B1 mutational status of primary MDS patient cells. Additional unimodal chromatin accessibility analysis showed increased priming of SF3B1 hematopoietic progenitors toward the megakaryocyte-erythroid lineage, as well as the enrichment of motifs from the TEA (TEAD) domain in accessible regions linked to genes with upregulated expression. Overall, chapter 2, applies a multi-stage fusion approach on transcriptomic data views to prioritize mis-spliced gene targets, and concurrently provides a formal overview of the SF3B1-mutated chromatin landscape and nominates transcriptional programs with putative roles in MDS disease biology. Chapter 3 examines if single cell gene expression signals together with the computational capacity of neural networks are able to predict a cell’s malignant status and subsequently its genotype for specific abnormalities in IDH1/2-mutated Acute Myeloid Leukemia (AML). To this end, using single cell RNA sequencing data from 50,026 cells, a feedforward neural network was trained to predict the cell’s malignant or wild-type (WT) status in a binary fashion, achieving an accuracy of 98% on the holdout test set. Furthermore, in a multi-label setting, this work deploys a similar architecture to predict the mutational status of specific genomic abnormalities at the single cell level, showing a macro-average AUC ROC=0.84 and NRAS mutational status prediction AUC ROC=0.83 on the holdout test-set. Altogether, chapter 3 applies deep learning in a supervised context to explore the connection between single cell gene expression profiles and genotypes in IDH1/2 AML and shows the potential of such modeling approaches in capturing meaningful genotype-phenotype relationships.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Asimomitis, Georgios
- Ασιμομήτης, Γεώργιος
Subjects
dc:subject × 21- Υπολογιστική βιολογία
- Μηχανική μάθηση
- Σύνδεση δεδομένων
- Μυελογενή νεοπλάσματα
- Ομικά προφίλ
- Computational biology
- Machine learning
- Multi-view integration
- Myeloid neoplasms
- Omics data
- Βιολογία
- Φυσικές Επιστήμες
- Μαθηματική και Υπολογιστική βιολογία
- Βιοϊατρική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Biological Sciences
- Natural Sciences
- Mathematical and Computational Biology
- Medical Engineering
- Engineering and Technology
- Biomedical Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/56745