2006 Scientific Session Abstracts
Scientific Session Presentations & Schedule
E-Poster Presentations
Submission Compliance and Guidelines
Copyright Transfer Form (PDF)
Abstract Archive
An Edge Detection-Based Algorithm for
the Segmentation of Array CGH Data
Anna Chu1 (achu@bccrc.ca), Bradley P. Coe, BSc; Raj Chari, BSc1; Calum MacAulay, PhD1; Wan L. Lam, PhD1; 1British Columbia Cancer Research Centre, Vancouver, BC and University of British Columbia, Vancouver, BC
Context: Array Comparative Genomic Hybridization (aCGH) is a standard DNA microarray technique which allows for the detection of genomic imbalances. Current high resolution approaches such as tiling array bacterial artificial chromosome (BAC) aCGH, which offers full overlapping coverage of the genome with over 27,000 data points, allows detection of segmental genomic alterations as small as 80kb in size.
Segmentation and breakpoint detection algorithms allow for the systematic delineation of copy number gains and losses. Although a number of these algorithms exist, none of these algorithms have been designed to account for the additional information available from the overlapping elements present in whole genome tiled aCGH data.
Technology: The algorithm was implemented in Java and utilizes the statistical package R and is part of the SIGMA (System for Genomic Microarray Analysis) Viewer application.
Design: Described is a breakpoint detection algorithm that uses a subclone-Gaussian smoother and Canny edge detector to identify copy number breakpoints in tiling aCGH data. Copy number gain or loss is determined through statistical comparison of each breakpoint-derived segment against a series of stretches in the genome deemed to be the most “normal“. This representation of “normal” is comprised of sets of contiguous clones, which are located throughout the genome with an average Log2 ratio near 0 with minimal variance. Segments which do not test as statistically different from these regions of normal copy number are annotated to be copy number neutral, while other regions demonstrating significant average Log2 ratio deviations, either above or below normal, are annotated gains and losses respectively. To test the ability of this algorithm, we applied this to a previously analyzed dataset of lung cancer cell lines.
Results: Applying this algorithm to a set of lung cancer cell lines has identified both known and potentially novel regions of DNA aberration, which were not identified by other algorithms. In addition, data produced by this algorithm better agrees with visual confirmation by expert array analysts as well as experimental validation.
Conclusion: This edge detection based algorithm offers a novel breakpoint detection method to systematically annotate copy number changes using tiling array CGH data.