Paterson Institute for Cancer Research SCIENTIFIC REPORT 2005

8
GROUP LEADER
Crispin Miller
Bioinformatics Group
http://www.paterson.man.ac.uk/groups/bioinf.jsp
Modern molecular biology generates enormous data sets
that are too big to analyse by hand. Instead, computers must
be used to sift through the information and find the relevant
patterns and signals in the data. Bioinformatics is the study
of how these programs must be written and applied to biological
data. Our own research is focused on developing
novel techniques and software tools for analysing microarray
data and, increasingly, quantitative proteomics arising from
iTRAQ based mass spectrometry.
Both microarrays and high throughput quantitative
proteomics measure the expression of many thousands
of genes in parallel, raising issues of data
management, statistics and computing as well as
biology and biochemistry. Successful analysis relies
on understanding how each of these contributes to
the data produced by an experiment. In addition,
each sample can result in large amounts of data.
Recent Affymetrix chips, for example, use ~500,000
features to probe for ~54,000 different transcripts;
the simplest experiment comparing between two
samples in triplicate generates data for about
3,000,000 features at once. A clinical study might
involve hundreds of samples, and generate many
millions of data points for further analysis.
Data management and analysis
Microarray analysis relies on the use of statistical
tests to assess the significance of each change in
gene expression; experiments are repeated a number
of times to generate replicates, and the replicate
data used to evaluate the consistency of the
observed differences. These tests are often accompanied
by calculations of fold-change, produced
POSTDOCTORAL
FELLOWS
Michal Okoniewski
Claire Wilson
RESEARCH
APPLICATIONS
PROGRAMMER
Tim Yates
SYSTEM
ADMINISTRATOR/
SCIENTIFIC
PROGRAMMER
Zhi Cheng Wang
GRADUATE STUDENTS
Laura Edwards
(nee Hollins, with Lez Fairbairn)
Graeme Smethurst
(with Peter Stern)
from the mean values for each set of samples. The
majority of our data analysis work uses
BioConductor (www.bioconductor.org), a collection
of analysis tools built with the statistical programming
language R. We contribute code to
BioConductor and have also been continuing to
develop our own package, ‘simpleaffy’, which
implements a variety of analysis algorithms for
Affymetrix data, including Quality Control, signal
detection, expression level generation and a set of
graph plotting and visualisation functions, and
‘plier’, which uses a wrapper around Affymetrix’s
SDK to provide access to their ‘plier’ algorithm.
Knowledge of the replicate structure of a microarray
experiment is fundamental to its correct interpretation.
We have developed a large MIAME
compliant database that provides access to expression
data via a Web interface. In order to allow the
database to be searched for experiments in which
specified genes are differentially expressed, the
database must have access to information describing
the replicate structure of each experiment, and
use this to guide the statistical tests that underpin
the search. We have developed an annotation system
that uses a ‘drag-and-drop’ interface that allows
users to build a pictorial representation of their
experiment using a set of icons that represent the
different stages of the experimental process. The
system makes use of this apparently informal interaction
to build a structured, and machine readable
representation of experimental design. This is subsequently
used by the database to group samples
together to support a variety of tasks including data
visualization and gene-centred searches. As
datasets become larger and more complex to man-
P A T E R S O N I N S T I T U T E S C I E N T I F I C R E P O R T 2 0 0 5