Outliers when Clustering Microarray Data

Start: 03/26/2009 - 4:15pm
End  : 03/26/2009 - 5:15pm

Statistics/OR/Math Finance Seminar

Jo Hardin (Pomona)


Microarray data are well known to be noisy and rife with outliers. The outliers are sometimes interesting in their own right, but often they are simply poor quality measurements that should be removed from the analysis. Unlike many other statistical techniques, clustering methods will always give you cluster outputs regardless of the structure of the data. Though clustering results can be enormously informative, the results can also be misleading if the data have outlying values. In particular, when clustering genes with only tens of samples, a few outlying values can easily change the direction of the relationship between a pair of genes. We provide mechanisms for robust clustering that minimize unwanted noise. No background in microarrays or clustering needed for this talk.

Beckman B126, Harvey Mudd College