UNIVERSITY OF HERTFORDSHIRE
		 COMPUTER SCIENCE RESEARCH COLLOQUIUM


	   "DISCRIMINATING CODING, NON-CODING AND REGULATORY REGIONS
		IN GENOMES USING RESCALED RANGE AND DETRENDED
			  FLUCTUATION ANALYSIS"


			 Dr. Rene te Boekhorst
		       School of Computer Science
		       University of Hertfordshire

			7 December 2005 (Wednesday)

			  Lecture Theatre E350
		       Hatfield, College Lane Campus
			       3 - 4 pm

	       Coffee/tea and biscuits will be available.
			 [Catering Permitting]

		    Everyone is Welcome to Attend
			  [Space Permitting]


Abstract:

   In this talk I will explain some of the methods we use to characterize
and differentiate two important constituents of the genome: regulatory
regions and exons.
   Exons are those sequences of nucleotides (nucleotides are the
building blocks of the DNA molecule) that are involved in linking
up amino-acids, which in turn are the fundamental components of
proteins. Together with pieces of DNA that separate exons from each
other ('introns'), exons form genes. Proteins play various roles in the
life of organisms. They may have structural functions (such as keratin,
the building material of nails and hair) or act as regulators. Apart
from their (auto-) catalytic role as enzymes in metabolism and
development, a very important function of some regulatory proteins
is the control of gene expression. A particular class of these
expression regulating proteins, the so-called transcription factors,
currently attracts much research interest. By binding to certain
regions of the genome (the transcription factor binding sites) these
proteins determine which genes are activated or inhibited and at what
time. Transcription factor binding sites are clustered in so-called
regulatory regions and much of current bioinformatics research is
devoted to the development of tools that recognize these regions.
   The algorithms applied to identifying the position and function of
genomic elements can broadly be categorized as supervised and
non-supervised methods.  Supervised methods are based on a training
set of data (nucleotide sequences of known functional elements, such
as exons or regulatory regions) from which machine learning methods
are to abstract recognizable patterns. Non-supervised techniques
start without such knowledge, but use for instance the over- (or
under-) representation of certain sequences to infer about their
possible function.
   The methods we have been working on are a little bit of both. We used
sequences of known regulatory regions, exons and non-regulatory regions
(not exons) to search for statistical properties that discriminate
between these three types of genomic elements. The properties I
will focus on are entropy and sequential `persistence' (long range
dependency in the sequence of nucleotides) and I will show statistical
evidence for the differentiation these measurements allow for. If
time permits, I will also discuss how these statistics could be used
in a sliding window technique (in which the size of the window adapts
to the local structure of the DNA) to localize exons and binding sites.

This is joint work of the speaker with Irina Abnizova (MRC-BSU
Cambridge) and Chrystopher Nehaniv (Computer Science, University of
Hertfordshire).

--
Hertfordshire Computer Science Research Colloquium
http://homepages.feis.herts.ac.uk/~nehaniv/colloq