Graph-based and alignment-free analysis of bacterial pangenomes
File(s)
Author(s)
Horsfield, Samuel
Type
Thesis
Abstract
The collective genetic content of a population, defined as its ‘pangenome’, is much greater
than that of a single genome for many bacterial species. Over the last two decades, work has
focused on how best to represent pangenomes, enabling characterisation of the evolutionary
forces driving their formation and maintenance. In this thesis, I explore the applications of two
approaches to representing pangenomes, graph-based and alignment-free methods, to improve
the accuracy and computational efficiency of conventional bioinformatics practices.
Firstly, I developed a graph-based bacterial gene prediction and annotation tool, ggCaller, to
address inconsistency and computational redundancy issues occurring when annotating genomes
individually. I detail ggCaller development, benchmarking it on simulated and real bacterial
genome datasets. I show that ggCaller improves both accuracy and runtime of pangenome
analysis against current state-of-the-art tools.
Secondly, I explored the application of Nanopore Adaptive Sampling (NAS), a read-alignment
method for targeted DNA enrichment, to studying pangenome diversity within mixed samples.
I show NAS performs poorly when closely-related contaminant DNA is present in a sample,
remedied by targeting structurally-diverse species-specific loci. I then developed a graph-based
alignment-free method for read alignment, which improves NAS accuracy when targeting
unobserved variants of structurally-diverse loci.
Finally, I developed approaches for modelling pangenome evolution using the alignment-free
method, PopPUNK. PopPUNK estimates genome similarity using sequence matching, which
is highly scalable and insensitive to gene prediction errors, improving upon conventional genebased
modelling approaches. I show that PopPUNK data can distinguish species with distinct
population structures, and benchmark analytical and simulation-based methods which enable
study of evolutionary forces acting upon pangenomes.
This work provides the research community with publicly available high-accuracy tools
for pangenome analysis. It also highlights the benefits and challenges of graph-based and
alignment-free methods, which have the computational efficiency necessary to study pangenome
diversity in huge genome datasets.
than that of a single genome for many bacterial species. Over the last two decades, work has
focused on how best to represent pangenomes, enabling characterisation of the evolutionary
forces driving their formation and maintenance. In this thesis, I explore the applications of two
approaches to representing pangenomes, graph-based and alignment-free methods, to improve
the accuracy and computational efficiency of conventional bioinformatics practices.
Firstly, I developed a graph-based bacterial gene prediction and annotation tool, ggCaller, to
address inconsistency and computational redundancy issues occurring when annotating genomes
individually. I detail ggCaller development, benchmarking it on simulated and real bacterial
genome datasets. I show that ggCaller improves both accuracy and runtime of pangenome
analysis against current state-of-the-art tools.
Secondly, I explored the application of Nanopore Adaptive Sampling (NAS), a read-alignment
method for targeted DNA enrichment, to studying pangenome diversity within mixed samples.
I show NAS performs poorly when closely-related contaminant DNA is present in a sample,
remedied by targeting structurally-diverse species-specific loci. I then developed a graph-based
alignment-free method for read alignment, which improves NAS accuracy when targeting
unobserved variants of structurally-diverse loci.
Finally, I developed approaches for modelling pangenome evolution using the alignment-free
method, PopPUNK. PopPUNK estimates genome similarity using sequence matching, which
is highly scalable and insensitive to gene prediction errors, improving upon conventional genebased
modelling approaches. I show that PopPUNK data can distinguish species with distinct
population structures, and benchmark analytical and simulation-based methods which enable
study of evolutionary forces acting upon pangenomes.
This work provides the research community with publicly available high-accuracy tools
for pangenome analysis. It also highlights the benefits and challenges of graph-based and
alignment-free methods, which have the computational efficiency necessary to study pangenome
diversity in huge genome datasets.
Version
Open Access
Date Issued
2023-09
Date Awarded
2023-12
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Croucher, Nicholas
Lees, John
Grassly, Nicholas
Sponsor
Medical Research Council (Great Britain)
Great Britain. Foreign, Commonwealth and Development Office
Grant Number
MR/S502388/1
Publisher Department
School of Public Health
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)