Algorithms for Detecting Gene Copy Number Fluctuations in Tumor Cells
by Microarrays
________________________________________________________

Bud Mishra
________________________________________________________
Professor of Computer Science & Mathematics (Courant, NYU)
Professor of Bioinformatics (Adjunct, Cold Spring Harbor)
Professor of Human Genetics (Adjunct, Mt Sinai School of Medicine)

Hybridization, a process by which two complementary DNA strands match
by sequence specific base-pairing and form a double-stranded DNA
complex, has become a fundamental tool of genomics as it permits the
usage of DNA molecules themselves as the perfect reagent to identify
particular DNA sequences. Recently, microarrays have allowed many such
hybridization experiments to proceed concurrently, thus promising high
throughput.

A key biological problem in this context can be abstracted to a
graph-theoretic problem concerned with embedding its vertices on a
real line in such a manner that the pair-wise distance measurements
among the vertices are "preserved." We study a wide class of
distributions for the distance measurements and suggest several
algorithms to solve it with good accuracy and efficiency, i.e.,
O(E log V + V^2). We give a probabilistic analysis of the case when
the distance metric is inferred from a set of microarray experiments
using short probes hybridized to pooled fluorochrome labeled clones
from a large library.

We shall also explore various limitations of this technology
due to several error sources (noisy signal detection, imperfect
base-pairing, non-specific cross-hybridization, etc.) and suggest ways to
overcome them via algorithmic and statistical means. In particular, we   
shall focus on two specific applications: 1) Rapid detection of gene
amplifications (involving oncogenes) and gene deletion (involving   
tumor suppressor genes) in cancerous tumor cells, 2) Rapid and      
accurate placement of low-complexity probes along the genome to     
characterize gene amplification and deletions.

(Jointly with CSHL's Wigler-Lab, MSKCC's Larry Norton
and CIMS's Will Casey.)