Degree
Master of Science
Program
Computer Science
Supervisor
Dr. Lucian Ilie
Abstract
The study of protein-protein interactions (PPI) is critically important within the field of Molecular Biology, as proteins facilitate key organismal functions including the maintenance of both cellular structure and function. Current experimental methods for elucidating PPIs are greatly hindered by large operating costs, lengthy wait times, as well as low accuracy. The recent development of computational PPI predicting techniques has worked to address many of these issues. Despite this, many of these methods utilize over-engineered features and naive learning algorithms. With the recent advances in Machine Learning and Artificial Intelligence, we attempt to view this problem through a novel, deep learning perspective. We propose a siamese, convolutional neural-network architecture for predicting protein-protein interactions using protein signatures as feature vectors. In comparison to four leading computational methods, we find that our results are comparable to and, in many cases, surpass the results of these methods. The emphasis of the discussion is to show that there is still much room for improvement in the area of PPI prediction using modern deep learning techniques.
Recommended Citation
Ahmed, Muhammad S., "SIGNET: A Neural Network Architecture for Predicting Protein-Protein Interactions" (2017). Electronic Thesis and Dissertation Repository. 4889.
https://ir.lib.uwo.ca/etd/4889