Genetic algorithm learning: A new approach for the discovery of small, diffuse patterns in genomic sequences
Speaker: Shuba Gopal , Formerly with the Rochester Institute of Technology
Date: February 18 2009
Time: 11:30AM to 1:00AM
Location: Stata Center 32-G575
Host: Bonnie Berger & Peter Clote, MIT - BC
Contact: Patrice Macaluso, 617.253.3037, macaluso@csail.mit.edu
Relevant URL: http://www-math.mit.edu/compbiosem/
Genetic algorithm learning: A new approach for the discovery of small, di use patterns in genomic sequences
Genomic sequence analysis has centered on the identi cation of critical signals associated with genes and regulatory regions. A variety of methods, based on statistical inference and machine learning, have been used to detect the presence of genes, promoter signals
and associated patterns. However, such approaches have limitations. In particular, they generally require the pattern of interest to be easily discernible from random sequence noise
by at least one characteristic, e.g. distinctive sequence composition, consistent length or well-de ned locations within the genome. For example, the TATA box promoter sequence
has a distinctive composition, TATAA, a generally invariable length of ve nucleotides and is almost always located upstream of protein-coding eukaryotic genes. However, not all
signals involved with the regulation of genes are characterized by such distinctive features.
The signals associated with alternative splicing tend to be very short patterns (as few as 2 nucleotides), have weak sequence conservation and their locations within genes can be highly
variable. Yet it is precisely these types of signals that we must uncover as we learn how genes are manipulated to generate organismal complexity.
Given the limitations of pattern nding algorithms, a di erent approach has been pro-posed that uses genetic algorithms to better de ne patterns of interest. Genetic algorithms mimic biological evolution and have been applied with great success to solve a wide variety
of search and optimization tasks. By transforming the pattern nding challenge in genomes into a search and optimization task, genetic algorithms o er a powerful new way to identify
the features associated with a given regulatory signal. We present the genetic algorithm approach and describe one application to the discovery of the signals associated with an unusual regulatory process known as RNA editing.
James Thompson and Shuba Gopal. (2006). "Genetic algorithm learning as a robust approach to RNA editing site prediction," BMC Bioinformatics, 7:145. Correction in BMC Bioinformatics,
7:406.
See other events that are part of Bioinformatics Seminar Series Spring 2009
See other events happening in February 2009
