Abstract

Metabolic pathway reconstruction is the essence of systems biology where in silico modeling
and prediction of the cell's function is based on the interaction of the cell's components
represented as a network of reactions. The reconstructed model and the associated database
of information about the organism's genes and their functional roles facilitate a variety of
analysis and simulation techniques that can enrich our understanding. However, there are
unresolved issues for genome-scale metabolic network reconstruction, such as our incomplete
knowledge of the cell's networks for metabolism, transport, and regulation; the completeness,
accuracy, and specificity of the annotation of genomes; and our ability to fully utilise the
available information from -omics (genomics, proteomics, metabolomics, etc) for the reconstruction
of the networks. These issues result in incomplete metabolic models, which limit
our ability to perform analysis of and to make predictions about the cell that are based on
the network model.
This dissertation discusses the state-of-the-art of metabolic pathway reconstruction and highlights
the outstanding issues. In particular, we consider a number of case studies using
genomes of fungi relevant to industrial applications, such as biofuels, to demonstrate the
performance of existing techniques and illustrate the issues. Our case studies focus on the
cell's central metabolism, and the utilisation and transport of sugars as a carbon source,
since these are essential concerns for industrial applications.
A significant deficiency in the existing state-of-the-art for the reconstruction of metabolic
pathways is the ability to associate genes and proteins to the transport reactions that move
specific compounds across the membranes of the cell. The dissertation reviews the state-of-the-
art of prediction methods for transmembrane transport proteins by developing a scheme
to describe and compare existing methods, and applying the existing techniques to the
v
fungal genome of A. niger CBS 513.88. This reveals the split between those methods that
use the Transporter Classification (TC) as their target for prediction, and those that use
the type of chemical substrates being transported as their target. Despite this difficulty in
comparing approaches, it is clear that the state-of-the-art cannot predict specific substrates
being transported, and hence cannot associate genes and proteins to the transport reactions.
The dissertation presents TransATH, which stands for Transporters via ATH (Annotation
Transfer by Homology), a system which automates Saier's protocol and includes the computation
of subcellular localization and improves the computation of transmembrane segments.
The choice of thresholds for the parameters of TransATH is investigated to determine optimal
performance as defined by a gold standard set of transporters and non-transporters from
S. cerevisiae. The dissertation demonstrates TransATH on the fungal genome of A. niger
CBS 513.88 and evaluates the correctness of TransATH using the curated information in
AspGD (the Aspergillus Database). A website for TransATH is available for use.