Annual Scientific Report 2015

Goldman Group
Evolutionary Tools for
Genomic Analysis
The diversity of all life has been shaped by its evolutionary history. Our research
focuses on the processes of molecular sequence evolution, developing data
analysis methods that allow us to exploit this information and glean powerful
insights into genomic function, evolutionary processes and phylogenetic history.
To understand the evolutionary relationships between
all organisms, it is necessary to analyse molecular
sequences with consideration of their underlying
structure. This is usually represented by an evolutionary
tree indicating the branching relationships of organisms
as they diverge from their common ancestors, and
showing degrees of genetic difference between them.
We develop mathematical, statistical and computational
techniques to reveal information from genome data,
draw inferences about the processes that gave rise to
these interrelationships and make predictions about
the biology of the systems whose components are
encoded in those genomes. We develop new evolutionary
models and methods, sharing them via stand-alone
software and web services, and apply new techniques
to interesting biological questions. We participate in
comparative genomic studies, both independently and
in collaboration with others. Our evolutionary studies
involve the analysis of next-generation sequencing
(NGS) data, which enables enormous gains in our
understanding genomes but poses many new challenges.
Major achievements
In 2015 we investigated the impact of popular multiple
sequence alignment (MSA) programs on reconstruction
of ancestral sequences. Many researchers are interested
in synthesising proteins of parental or extinct species
to study their biochemical properties and compare
them with those of their extant relatives. Accurate
reconstruction of ancestral states is vital to these
studies; however, we discovered that different aligners
introduce various biases. We also studied the impact of
different models on estimation of divergence time, as
traditional substitution models tend to underestimate
genetic distances between distantly related species,
introducing large biases. By relaxing the assumption
of site-invariant selective pressures, we demonstrated
that longer distances are estimated for basal branches of
species trees.
We published the results of a long-running comparative
study of MSA filtering methods, which automatically
identify and remove unreliable alignment sites,
including data that may introduce errors into
evolutionary analyses. The filtering step has been
assumed to lead to better trees; however, our study
showed that with most filtering methods and in most
circumstances, alignment filtering worsens the resulting
trees. These findings have implications for scientists
working with difficult-to-align sequences.
We investigated how the presence of gaps in a MSA – for
example those introduced by insertions or deletions
of genetic material (indels) – affects the accuracy of
inferred phylogenies. Standard phylogenetic methods
do not attempt to model indels, rather treating gaps
as missing data. To address the suggestion that this
could lead to statistical inconsistency giving rise to an
incorrect evolutionary tree, even as more and more
data are collected, we derived a new, simple proof
of statistical consistency of maximum likelihood
phylogenetic reconstruction for un-gapped alignments.
In so doing we showed that the suggested inconsistency
only pertains to one, very specific, outlier case.
As most cell divisions introduce a small number
of mutations into the genome, it is possible to use
single-cell exome/genome sequencing data to infer
somatic evolutionary histories of individual cells.
These phylogenies, called ‘cell lineage trees’, are useful
in developmental biology and cancer research but
are difficult to resolve due to very low mutation rates
and high sequencing error rates resulting from allelic
dropout. The most commonly used algorithm does
not distinguish between mutations and sequencing or
sample amplification errors, so we developed a scalable
algorithm based on a phylogenetic model that explicitly
models mutations and sequencing errors as two separate
processes. In addition to resolving a long-standing,
tractable, mathematical problem, the new method gives
more accurate trees than those produced using standard
existing methods.
‘Incongruence’, increasingly common with the rise of
multi-locus and whole-genome sequencing, occurs
when an estimated tree varies depending on the
particular set of sequences on which it was built.
Incongruence can arise due to stochastic differences
between loci, or to different sets of sequences having
undergone different processes of evolution. To address
the latter case we developed treeCl, a clustering method
that groups loci that share a common evolutionary
history and distinguishes sets that do not. We tested
the method using both simulated data, a curated
set of yeast proteins and RAD-sequenced DNA loci
from globeflower flies (Chiastocheta spp.; Diptera,
Anthomyiidae). We improved the curation of the yeast
data by identifying several non-orthologous sequences,
137
2015 EMBL-EBI Annual Scientific Report