at 14:00
Tel-Aviv University - Computer Science Colloquium
Sunday, October
31, 14:15-15:15
at 14:00
Room 309
Schreiber Building
Different subsets of an organism's total gene pool are expressed when
cells undergo different biological processes. Identifying the
particular genes expressed at a given stage of a process and measuring
their relative abundance can help the characterization of gene function
in different conditions. Recently developed DNA arrays enable
simultaneous measurements of the expression levels of thousands of
genes. These methodologies have led to a tremendous acceleration in
the
rate at which gene expression data is accumulated. The large size of
the data, the complexity of the underlying processes, and the inherent
high noise level present a major challenge for analysis.
In the first part of the talk I will present clustering algorithms
developed for the analysis of gene expression data. On the theoretical
side, we present an appropriate stochastic model for the data, called
the {\em planted partition model}. We present a fast clustering
algorithm that recovers the planted partition with high probability.
On
the practical side, a heuristic version of the algorithm was
implemented (in the BioClust Package). This package is routinely used
to analyze data, and examples will be presented.
Gene expression data is expected to provide insight into cancer related
cellular processes, and to serve as a basis to diagnostic platforms.
Toward this end, several recent experiments measured gene expression
in
samples from tumor and normal tissues. In the second part of the talk
I
will present a novel clustering based classification method. We use
this algorithm to assess the cancer classification power of gene
expression data. We present results of performing leave-one-out cross
validation (LOOCV) experiments on two data sets, derived from colon
cancer and ovarian cancer studies. We demonstrate success rate of at
least 90\% in tumor vs. normal classification.
The talk will be self contained.
For colloquium schedule, see http://www.math.tau.ac.il/~zwick/colloq.html