Journal Article

Accurate confidence aware clustering of array CGH tumor profiles

Bart P.P. van Houte and Jaap Heringa

in Bioinformatics

Volume 26, issue 1, pages 6-14
Published in print January 2010 | ISSN: 1367-4803
Published online October 2009 | e-ISSN: 1460-2059 | DOI:
Accurate confidence aware clustering of array CGH tumor profiles

More Like This

Show all results sharing this subject:

  • Bioinformatics and Computational Biology


Show Summary Details


Motivation: Chromosomal aberrations tend to be characteristic for given (sub)types of cancer. Such aberrations can be detected with array comparative genomic hybridization (aCGH). Clustering aCGH tumor profiles aids in identifying chromosomal regions of interest and provides useful diagnostic information on the cancer type. An important issue here is to what extent individual aCGH tumor profiles can be reliably assigned to clusters associated with a given cancer type.

Results: We introduce a novel evolutionary fuzzy clustering (EFC) algorithm, which is able to deal with overlapping clusterings. Our method assesses these overlaps by using cluster membership degrees, which we use here as a confidence measure for individual samples to be assigned to a given tumor type. We first demonstrate the usefulness of our method using a synthetic aCGH dataset and subsequently show that EFC outperforms existing methods on four real datasets of aCGH tumor profiles involving four different cancer types. We also show that in general best performance is obtained using 1− Pearson correlation coefficient as a distance measure and that extra preprocessing steps, such as segmentation and calling, lead to decreased clustering performance.

Availability: The source code of the program is available from


Supplementary information: Supplementary data are available at Bioinformatics online.

Journal Article.  6337 words.  Illustrated.

Subjects: Bioinformatics and Computational Biology

Full text: subscription required

How to subscribe Recommend to my Librarian

Users without a subscription are not able to see the full content. Please, subscribe or login to access all content.