A comprehensive sensitivity analysis of microarray breast cancer classification under feature variability
Large discrepancies in signature composition and outcome concordance have been observed between different microarray breast cancer expression profiling studies. This is often ascribed to differences in array platform as well as biological variability.
We conjecture that other reasons for the observeddiscrepancies are the measurement error associated with each feature and the choice of preprocessing method. Microarray data are known to be subject to technical variation and the confidence intervals around individual point estimates of expression levels can be wide.
Furthermore, the estimated expression values also vary depending on the selected preprocessing scheme. In microarray breast cancer classification studies, however, these two forms of feature variability are almost always ignored and hence their exact role is unclear.
Results: We have performed a comprehensive sensitivity analysis of microarray breast cancer classification under the two types of feature variability mentioned above.
We used data from six state of the art preprocessing methods, using a compendium consisting of eight different datasets, involving 1131 hybridizations, containing data from both one and two-color array technology. For a wide range of classifiers, we performed a joint study on performance, concordance and stability.
In the stability analysis we explicitly tested classifiers for their noise tolerance by using perturbed expression profiles that are based on uncertainty information directly related to thepreprocessing methods. Our results indicate that signature composition is strongly influenced by feature variability, even if the array platform and the stratification of patient samples are identical.
In addition, we show that there is often a high level of discordance between individual class assignments for signatures constructed on data coming from different preprocessing schemes, even if the actual signature composition is identical.
Conclusions: Feature variability can have a strong impact on breast cancer signature composition, as well as the classification of individual patient samples. We therefore strongly recommend that feature variability is considered in analyzing data from microarray breast cancer expression profiling experiments.
Author: Herman SontropPerry MoerlandRene van den HamMarcel ReindersWim Verhaegh
Credits/Source: BMC Bioinformatics 2009, 10:389
Published on: 2009-11-26
Copyright by the authors listed above - made available via BioMedCentral (Open Access). Please
make sure to read our disclaimer prior to contacting 7thSpace Interactive. To contact our editors, visit our online helpdesk. If you wish submit your own press release, click here.
Social Bookmarking
RETWEET This! | Digg this! | Post to del.icio.us | Post to Furl | Add to Netscape | Add to Yahoo! | Rojo
|
|
