PhD Comprehensive Examination Announcement of Shaghayegh Sadeghi:"Representation Learning Methods for Association Prediction Tasks in Pharmacology "

Monday, December 12, 2022 - 11:00 to 13:00


The School of Computer Science would like to present…   

PhD. Comprehensive Exam by: Shaghayegh Sadeghi 

Date: Monday December 12, 2022 
Time: 11:00 AM- 1:00PM 
Location: Essex Hall, Room 122 
Reminders: Two-part attendance mandatory, arrive 5-10 minutes prior to event starting - LATECOMERS WILL NOT BE ADMITTED once the door has been closed and the presentation has begun. Please be respectful of the presenter by NOT knocking on the door for admittance.


De-novo pharmacology has been transformed by computational drug discovery in the past decade. One of the main topics in computational drug discovery is association prediction which often will be cast as a link prediction problem. The objective of link prediction in this research is to identify missing links between pairs of nodes. A study on link prediction tasks on drug-related graphs will benefit many research fields, including Drug repurposing (DR), Drug target interaction prediction (DTI), and Drug-drug interaction prediction (DDI). Association data can be represented by a network. The input of this task are homogeneous networks such as; protein-protein, drug-drug, and disease-disease association networks, and heterogeneous networks such as; drug-disease, drug-side-effect, and protein-drug. Representation-based methods are able to learn from these networks by mapping nodes and edges into a latent space. The geometry of this latent space will show the structure of interaction between nodes and edges, and so that nodes and edges that are closer in the network are also geometrically closer in the latent space. This embedding information can be used as feature vectors for many downstream tasks such as node classification and link prediction. In this study, the focus is on the link prediction task. 
Therefore, an overview of the link prediction task is provided in this comprehensive study, which consists of four main sections. First, a study of traditional embedding methods using only graph-structured data on homogeneous graphs will be presented. Then, embedding heterogeneous graphs based on only structural data will be provided. Due to the nature of bioinformatics entities, which have rich side information, and the fact that classic embedding methods ignore them, deep embedding methods may provide a solution by integrating graph-structured data with node entities. Hence, in our third section, the focus is on node features (drug chemical structure) and the importance of using node content in link prediction tasks on the homogenous network (DDI network). And finally, a study of existing methods for the deep embedding of heterogeneous graphs (DR network, DTI network) will be discussed. 
Keywords: Link prediction, Representation learning, Graph neural network, Node Embedding 


PhD Doctoral Committee: 

External Reader: Dr. Mohammad Hassanzadeh 
Internal Reader: Dr. Luis Rueda 
Internal Reader: Dr. Saeed Samet 
Advisor(s): Dr. Jianguo Lu, Dr. Alioune Ngom 



Vector Institute in Artificial Intelligence, artificial intelligence approved topic logo

5113 Lambton Tower 401 Sunset Ave. Windsor ON, N9B 3P4 (519) 253-3000 Ext. 3716