MSc Thesis Proposal " DeepGLIOMarker: A Novel Framework for Glioblastoma Biomarker Identification from Protein Interaction Network Using Natural Language Processing and Graph Convolutional Networks" By: Zannatul Ferdoush

DeepGLIOMarker: A Novel Framework for Glioblastoma Biomarker Identification from Protein Interaction Network Using Natural Language Processing and Graph Convolutional Networks

MSc Thesis Proposal by:
Name: Zannatul Ferdoush

Date: Tuesday, January 23, 2024
Time: 12:00 pm – 1:00 pm
Location: Essex Hall Room 105

Abstract:
Glioblastoma multiforme (GBM), the most aggressive form of brain tumour, presents considerable challenges in oncology due to its complex nature and poor prognosis. In this study, we introduce DeepGLIOMarker, a novel computational framework that synergies natural language processing (NLP) and graph convolutional networks (GCNs) to identify potential GBM biomarkers from protein-protein interaction (PPI) networks. Our approach begins with the analysis of microarray datasets (GSE15824 and GSE16011) from the Gene Expression Omnibus (GEO) database, leading to the identification of differentially expressed genes (DEGs) between GBM and non-tumour tissues. These DEGs are integrated into a high-confidence PPI network using the STRING online tool and Cytoscape software. To capture the complex biological relationships between genes, we employ NLP techniques, specifically Word2Vec, to encode Gene Ontology (GO) annotations. GO annotations are crucial as they provide comprehensive information about gene functions, biological processes, and cellular components, enriching our understanding of how genes interact within the network. This encoding reveals semantic relationships between genes, providing a rich, feature-enriched representation of each node in the PPI network. Using multi-layered GCNs further enables the effective learning of gene representations in the context of the PPI network, facilitating the accurate classification of potential GBM biomarkers. Validation through survival analysis showed a significant link between identified biomarkers and patient survival underscoring their importance in GBM treatment. DeepGLIOMarker, combining NLP techniques and GCNs with survival and correlation analyses, offers new insights into GBM's molecular aspects, paving the way for potential therapies.

Keywords: Glioblastoma Multiform, Biomarker, Natural Language Processing, Graph Convolutional Networks

Thesis Committee:
Internal Reader: Dr. Dan Wu
External Reader: Dr. Mir Munir Rahim
Advisor: Dr. Ziad Kobti