Use this URL to cite or link to this record in EThOS:
Title: A network component analysis based divide and conquer method for transcriptional regulatory network analysis
Author: Prabhu Haladi Ramanatha, Sachin
ISNI:       0000 0004 7655 4261
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Understanding gene regulation has played a major role in several biomedical applications ranging from cancer studies to genetic engineering. Transcriptional regulatory networks (TRN) have been studied extensively to understand rules of interactions between transcription factor (TF) and genes that constitute gene regulation. In a particular type of gene expression data modelling problem, only gene expression profiles and regulatory patterns are available. Regulatory patterns or networks are binary matrices that indicate connections between genes and TFs. A TRN modelling method must simultaneously estimate regulatory strengths and concentrations of TFs. Therefore, TRN modelling problem is a structure-constrained matrix factorisation (SCMF) problem. In this thesis, among various available TRN modelling algorithms, Network Component Analysis (NCA) is chosen for further investigation. Several methods that extend NCA are proposed in literature. However, the following fundamental issues in NCA theory have remained unresolved 1. a method to test feasibility of NCA problem does not exist 2. a method to solve NCA problem with an infeasible start does not exist One of the major contributions in this thesis is a method to test NCA feasibility. It is made possible for the first time in relevant literature to test a datasetnetwork pair for NCA feasibility before applying NCA. This is done by translating NCA rank conditions on posteriori variables to rank conditions on a priori available dataset-network pair. In this process, it is shown that binary rank of a regulatory pattern is important to define NCA feasibility. Another major contribution in thesis is a divide and conquer method that computes unique solutions corresponding to NCA infeasible dataset-network pairs. Techniques that extend NCA to solve SCMF problems with an infeasible start proposed in literature are shown to be inaccurate and limited in applicability. In this thesis, a bipartite matching based approach is developed to decompose an infeasible NCA network into a set of full-rank factorisable sub-networks. A solution corresponding to the original network is obtained as a convex combination of matrix factors corresponding to identified sub-networks. It is shown that the resulting matrix factors for the whole network are unique up to two scaling factors if all data subset-subnetwork pairs are NCA feasible.
Supervisor: Wei, Hua-Liang Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available