Computational approaches to detecting and understanding changes in chromosome structure

Duplications and deletions of portions of DNA are relatively common in cells that undergo multiple divisions. Such genetic changes often serve as a mechanism of rapid adaptation and can lead to increased fitness in cell populations under selective pressure. Furthermore, numerous cancers have been associated with widespread abnormalities of this type that are believed to disrupt key pathways and lead to uncontrolled growth. Accurate and precise identifation of when and where duplications and deletions occur will be the key to understanding this important adaptive mechanism and a step toward effective cancer treatment.

Recent developments in microarray technology have enabled genome-wide investigations of DNA duplications and deletions. However, extracting the relevant information from these experiments depends largely on computational tools from machine learning and statistics. I will discuss an approach we've developed for identifying chromosomal duplications and deletions based on the expectation-maximization (EM) algorithm as well as several non-parametric statistical tools that are of general interest. Given these tools, an important extension is identifying which putative duplications or deletions are most functionally relevant (i.e. responsible for a particular phenotype). I will present one solution to this problem based on the fusion of data from multiple sources.